Está en la página 1de 302

Create your own

Customized Content
Bundlethe more
books you buy,
the greater your
discount!

Manufacturing
Engineering
Mechanical
& Chemical
Engineering
Materials Science
& Engineering
Civil &
Environmental
Engineering
Electrical
Engineering

THE TERMS
Perpetual access for
a one time fee
No subscriptions or
access fees
Unlimited
concurrent usage
Downloadable PDFs
Free MARC records
For further information,
a free trial, or to order,
contact:
sales@momentumpress.net

Steven OHara and Carisa H. Ramming


As structural engineers move further into the age of digital computation and rely more heavily on computers to solve problems,
it remains paramount that they understand the basic mathematics and engineering principles used to design and analyze building structures. The link between the basic concepts and application to real world problems is one of the most challenging
learning endeavors that structural engineers face.
The primary purpose of Numerical Structural Analysis is to
assist structural engineering students with developing the ability to solve complex structural analysis problems. This book will
cover numerical techniques to solve mathematical formulations,
which are necessary in developing the analysis procedures for
structural engineering. Once the numerical formulations are understood, engineers can then develop structural analysis methods that use these techniques. This will be done primarily with
matrix structural stiffness procedures. Finally, advanced stiffness
topics will be developed and presented to solve unique structural problems, including member end releases, non-prismatic,
shear, geometric, and torsional stiffness.
Steven OHara is a professor of architectural engineering and licensed engineer at the Oklahoma State University School of Architecture since 1988. Professor OHara is an affiliate professor
of civil engineering and architecture in Project Lead the Way,
a nationwide program for high-school students. He trains the
high school teachers responsible for introducing engineering
principles into the secondary curriculum, through project-based
learning and coauthored the workbook for the curriculum, Civil
Engineering & Architecture Workbook. He is also coauthor of
the ARE Review Manual.
Carisa H. Ramming is a graduate of Oklahoma State University
with a bachelor of architectural engineering degree and dual
masters degree: master of science in civil engineering with an
emphasis in construction engineering and master of architectural engineering. Professor Ramming teaches engineering mechanics and has been faculty at OSU since 2009. She recently
coauthored Civil Engineering & Architecture Workbook for
Project Lead the Way, which provides curriculum for science,
technology, engineering, and mathematics education used in
middle and high schools.
ISBN: 978-1-60650-488-8

Numerical Structural Analysis

THE CONTENT

Numerical Structural Analysis

OHARA RAMMING

EBOOKS
FOR THE
ENGINEERING
LIBRARY

SUSTAINABLE STRUCTURAL
SYSTEMS COLLECTION
Mohammad Noori, Editor

Numerical
Structural
Analysis

Steven OHara
Carisa H. Ramming

NUMERICAL
STRUCTURAL
ANALYSIS

NUMERICAL
STRUCTURAL
ANALYSIS

STEVEN E. OHARA
CARISA H. RAMMING

MOMENTUM PRESS, LLC, NEW YORK

Numerical Structural Analysis


Copyright Momentum Press, LLC, 2015.
All rights reserved. No part of this publication may be reproduced, stored
in a retrieval system, or transmitted in any form or by any means
electronic, mechanical, photocopy, recording, or any otherexcept for
brief quotations, not to exceed 400 words, without the prior permission
of the publisher.
First published by Momentum Press, LLC
222 East 46th Street, New York, NY 10017
www.momentumpress.net
Proudly sourced and uploaded by [StormRG]
Kickass Torrents | TPB | ET | h33t

ISBN-13: 978-1-60650-488-8 (print)


ISBN-13: 978-1-60650-489-5 (e-book)
Momentum Press Sustainable Structural Systems Collection
DOI: 10.5643/9781606504895
Cover and Interior design by Exeter Premedia Services Private Ltd.,
Chennai, India
10 9 8 7 6 5 4 3 2 1
Printed in the United States of America

Abstract
As structural engineers move further into the age of digital computation and rely more heavily on computers to solve problems, it remains
paramount that they understand the basic mathematics and engineering
principles used to design and analyze building structures. The analysis of
complex structural systems involves the knowledge of science, technology, engineering, and math to design and develop efficient and economical buildings and other structures. The link between the basic concepts and
application to real world problems is one of the most challenging learning
endeavors that structural engineers face. A thorough understanding of the
analysis procedures should lead to successful structures.
The primary purpose of this book is to develop a structural engineering students ability to solve complex structural analysis problems that
they may or may not have ever encountered before. The book will cover
and review numerical techniques to solve mathematical formulations.
These are the theoretical math and science principles learned as prerequisites to engineering courses, but will be emphasized in numerical formulation. A basic understanding of elementary structural analysis is important
and many methods will be reviewed. These formulations are necessary
in developing the analysis procedures for structural engineering. Once
the numerical formulations are understood, engineers can then develop
structural analysis methods that use these techniques. This will be done
primarily with matrix structural stiffness procedures. Both of these will
supplement both numerical and computer solutions. Finally, advanced
stiffness topics will be developed and presented to solve unique structural
problems. These include member end releases, nonprismatic, shear, geometric, and torsional stiffness.

KEY WORDS
adjoint matrix, algebraic equations, area moment, beam deflection, carry-over factor, castiglianos theorems, cofactor matrix, column matrix,
v

vi Abstract

complex conjugate pairs, complex roots, conjugate beam, conjugate pairs,


convergence, diagonal matrix, differentiation, distinct roots, distribution
factor, eigenvalues, elastic stiffness, enke roots, extrapolation, flexural
stiffness, geometric stiffness, homogeneous, identity matrix, integer, integration, interpolation, inverse, joint stiffness factor, linear algebraic equations, lower triangular matrix, matrix, matrix minor, member end release,
member relative stiffness factor, member stiffness factor, moment-distribution, non-homogeneous, non-prismatic members, partial pivoting,
pivot coefficient, pivot equation, polynomials, principal diagonal, roots,
rotation, rotational stiffness, row matrix, second-order stiffness, shear
stiffness, slope-deflection, sparse matrix, square matrix, stiffness matrix,
structural flexibility, structural stiffness, symmetric transformation, torsional stiffness, transcendental equations, transformations, transmission,
transposed matrix, triangular matrix, upper triangular matrix, virtual
work, visual integration

Contents
List of Figures

xi

List of Tables

xv

Acknowledgments

xix

1Roots of Algebraic and Transcendental Equations

1.1 Equations

1.2 Polynomials

1.3 Descartes Rule

1.4 Synthetic Division

1.5 Incremental Search Method

10

1.6 Refined Incremental Search Method

12

1.7 Bisection Method

13

1.8 Method of False Position or Linear Interpolation

15

1.9 Secant Method

17

1.10 NewtonRaphson Method or Newtons Tangent

18

1.11 Newtons Second Order Method

21

1.12 Graeffes Root Squaring Method

24

1.13 Bairstows Method

38

References

46

2Solutions of Simultaneous Linear Algebraic Equations


Using Matrix Algebra

47

2.1 Simultaneous Equations

47

2.2 Matrices

48

2.3 Matrix Operations

52

2.4 Cramers Rule

56

2.5 Method of Adjoints or Cofactor Method

57

vii

viiiContents

2.6 Gaussian Elimination Method

63

2.7 GaussJordan Elimination Method

68

2.8 Improved GaussJordan Elimination Method

72

2.9 Cholesky Decomposition Method

73

2.10 Error Equations

78

2.11 Matrix Inversion Method

80

2.12 GaussSeidel Iteration Method

86

2.13 Eigenvalues by Cramers Rule

87

2.14 FaddeevLeverrier Method

90

2.15 Power Method or Iteration Method

91

References

95

3Numerical Integration and Differentiation

97

3.1 Trapezoidal Rule

97

3.2 Romberg Integration

99

3.3 Simpsons Rule

104

3.4 Gaussian Quadrature

109

3.5 Double Integration by Simpsons


One-Third Rule

112

3.6 Double Integration by Gaussian Quadrature

114

3.7 Taylor Series Polynomial Expansion

116

3.8 Difference Operators by Taylor Series Expansion

118

3.9 Numeric Modeling with Difference Operators

123

3.10 Partial Differential Equation Difference Operators

136

3.11 Numeric Modeling with Partial Difference Operators

140

References

145

4Matrix Structural Stiffness

147

4.1 Matrix Transformations and Coordinate Systems

147

4.2 Rotation Matrix

148

4.3 Transmission Matrix 

153

4.4 Area Moment Method

155

4.5 Conjugate Beam Method

158

4.6 Virtual Work

161

4.7 Castiglianos Theorems

165

4.8 Slope-Deflection Method

168

Contents ix

4.9 Moment-Distribution Method

170

4.10 Elastic Member Stiffness, X-Z System

174

4.11 Elastic Member Stiffness, X-Y System

181

4.12 Elastic Member Stiffness, 3-D System

186

4.13 Global Joint Stiffness

187

References

204

5Advanced Structural Stiffness

205

5.1 Member End Releases, X-Z System

205

5.2 Member End Releases, X-Y System

214

5.3 Member End Releases, 3-D System

217

5.4 Non-prismatic Members

229

5.5 Shear Stiffness, X-Z System

239

5.6 Shear Stiffness, X-Y System

243

5.7 Shear Stiffness, 3-D System

247

5.8 Geometric Stiffness, X-Y System

251

5.9 Geometric Stiffness, X-Z System

258

5.10 Geometric Stiffness, 3-D System

260

5.11 Geometric and Shear Stiffness

262

5.12 Torsion

265

5.13 Sub-structuring

266

References

268

About the Authors

271

Index

273

List of Figures
Figure 1.1. Incremental search method.

11

Figure 1.2. Refined incremental search method.

12

Figure 1.3. Bisection method.

14

Figure 1.4. Method of false position or linear interpolation.

16

Figure 1.5.

Secant method. 

18

Figure 1.6.

NewtonRaphson method or Newtons tangent.

19

Figure 1.7. NewtonRaphson method or Newtons tangent.

20

Figure 1.8. NewtonRaphson method or Newtons tangent.

20

Figure 1.9. Newtons second order method.

21

Figure 1.10. Newtons second order method.

23

Figure 1.11. Graeffes root squaring method.

27

Figure3.1. Trapezoidal rule.

98

Figure3.2. Romberg integration.

100

Figure3.3. Simpsons rule.

104

Figure3.4. Example 3.10 Simple beam with difference operator. 124


Figure3.5. Example 3.10 Simple beam with difference operator. 124
Figure3.6. Example 3.10 Simple beam with difference operator. 125
Figure3.7. Example 3.11 Fixed beam with difference operator.

129

Figure3.8. Numeric modeling with difference operators.

132

Figure3.9. Example 3.12 Column buckling with difference


operator.

134

Figure3.10. Numeric modeling with partial difference operators. 140


Figure3.11. Example 3.19 Plate bending.

141

Figure3.12. Example 3.19 Plate bending.

142

Figure3.13. Example 3.19 Plate bending.

142

xi

xii List of Figures

Figure4.1. Coordinate systems.

148

Figure4.2. 
Example4.1 Rotation, a.

149

Figure4.3. Example4.2 Rotation, b.

150

Figure4.4. Example4.3 Rotation, g.

151

Figure4.5. Transformation locations.

153

Figure4.6. Orthogonal forces.

153

Figure4.7. Transformation effects.

154

Figure4.8. Example4.5 Area moment.

156

Figure4.9. Example4.5 Area moment.

156

Figure4.10. Example4.5 Area moment.

157

Figure4.11. Example4.5 Area moment.

157

Figure4.12. Conjugate versus real supports.

159

Figure4.13. Conjugate versus real beams.

159

Figure4.14. Example4.6 Conjugate beam.

160

Figure4.15. Example4.6 Conjugate beam.

161

Figure4.16. Example4.8 Virtual work.

162

Figure4.17. Example4.8 Virtual work.

162

Figure4.18. Example4.9 Visual integration.

163

Figure4.19. Example4.9 Visual integration.

164

Figure4.20. Example4.9 Visual integration.

164

Figure4.21. 
Example4.10 Castiglianos second theorem.

166

Figure4.22. Example4.10 Castiglianos second theorem.

166

Figure4.23. Example4.11 Castiglianos second theorem.

167

Figure4.24. Example4.12 Slope-deflection.

169

Figure4.25. Example4.13 Moment-distribution.

172

Figure4.26. Elastic stiffness.

175

Figure4.27. Example4.15 qiy stiffness.

176

Figure4.28. Example4.15 qiy stiffness.

176

Figure4.29. Example4.16 iz stiffness.

178

Figure4.30. Example4.17 qiz stiffness.

181

Figure4.31. Example4.18 iy stiffness.

183

Figure4.32. Example4.18 iy stiffness.

184

Figure4.33. Example4.19 Global joint stiffness.

190

List of Figures xiii

Figure4.34. Example4.19 Global joint stiffness.

191

Figure4.35. Example4.20 Global joint stiffness.

197

Figure5.1. Example5.1 iz end release.

206

Figure5.2. Example5.2 qiy end release.

209

Figure5.3. Example5.2 qiy end release.

210

Figure5.4. Example5.2 qiy end release.

211

Figure5.5. Example5.3 Member stiffness.

218

Figure5.6. Example5.4 Non-prismatic member stiffness.

230

Figure5.7. Example5.5 Non-prismatic member stiffness.

235

Figure5.8. Example5.6 Non-prismatic member stiffness.

236

Figure5.9. Example5.7 Non-prismatic member stiffness.

237

Figure5.10. Example5.8 Shear stiffness.

239

Figure5.11. Example5.9 Shear stiffness.

241

Figure5.12. Example5.10 Shear stiffness.

243

Figure5.13. Example5.11 Shear stiffness.

245

Figure5.14. Example5.12 Shear area.

249

Figure5.15. Example5.13 Shear area.

250

Figure5.16. Geometric stiffness.

251

Figure5.17. Example5.14 Geometric stiffness.

252

Figure5.18. Example 5.15 Geometric stiffness.

255

Figure5.19. Example5.16 Geometric stiffness.

258

Figure5.20. Example5.17 Geometric stiffness.

260

List of Tables
Table 1.1. Synthetic division

Table 1.2. Example 1.5 Synthetic division

Table 1.3. Example 1.5 Synthetic division

Table 1.4. Example 1.5 Synthetic division

Table 1.5. Example 1.5 Synthetic division

Table 1.6. Example 1.6 Synthetic division

Table 1.7. Example 1.6 Synthetic division

Table 1.8. Example 1.6 Synthetic division

10

Table 1.9. Example 1.6 Synthetic division

10

Table 1.10. Example 1.7 Incremental search method

11

Table 1.11. Example 1.8 Refined incremental search method

13

Table 1.12. Example 1.9 Bisection method

15

Table 1.13. Example 1.10 Method of false position

17

Table 1.14. Example 1.11 Secant method

18

Table 1.15. Example 1.12 NewtonRaphson method

21

Table 1.16. Example 1.13 Newtons second order method

24

Table 1.17. Graeffes root squaring method

26

Table 1.18. Example 1.14 Graeffes root squaring methodreal and


distinct roots
30
Table 1.19. Example 1.15 Graeffes root squaring methodreal and
equal roots
34
Table 1.20. Example 1.16 Graeffes root squaring methodreal and
complex roots
37
Table 1.21. Bairstows method

39

Table 2.1. Example 2.4 Cramers rule

58

xv

xvi List of Tables

Table 2.2. 
Example 2.5 Cofactor method

61

Table 2.3. 
Example 2.6 Method of adjoints

62

Table 2.4. Example 2.7 Gaussian elimination

65

Table 2.5. Example 2.8 Gaussian elimination

67

Table 2.6. Example 2.10 Improved GaussianJordan


elimination method

74

Table 2.7. Example 2.11 Cholesky decomposition method

77

Table 2.8. Example 2.11 Cholesky decomposition method

78

Table 2.9. Example 2.11 Cholesky decomposition method

78

Table 2.10. Example 2.11 Cholesky decomposition method

78

Table 2.11. Example 2.12 Error equations

81

Table 2.12. Example 2.12 Error equations

82

Table 2.13. Example 2.12 Error equations

83

Table 2.14. Example 2.13 Matrix inversion method

84

Table 2.15. Example 2.13 Matrix inversion method

85

Table 2.16. Example 2.13 Matrix inversion method

85

Table 2.17. Example 2.13 Matrix inversion method

86

Table 2.18. Example 2.14 GaussSeidel iteration method

88

Table 2.19. Example 2.16 FaddevLeverrier method

92

Table 2.20. Example 2.17 Power method

94

Table3.1. Example 3.1 Trapezoidal rule

98

Table3.2. Example 3.1 Trapezoidal rule

99

Table3.3. Example 3.2 Romberg integration

102

Table3.4. Example 3.2 Romberg integration

102

Table3.5. Example 3.2 Romberg integration

102

Table3.6. Example 3.2 Romberg integration

102

Table3.7. Example 3.3 Simpsons one-third rule

107

Table3.8. Example 3.3 Simpsons one-third rule

107

Table3.9. Example 3.4 Simpsons one-third and


three-eighths rules

108

Table3.10. Example 3.4 Simpsons one-third and


three-eighths rules

108

Table3.11. Gaussian quadrature

110

Table3.12. Example 3.5 Gaussian quadrature

110

Table3.13. Example 3.5 Gaussian quadrature

111

List of Tables xvii

Table3.14. Example 3.5 Gaussian quadrature

111

Table3.15. Example 3.6 Double integration by Simpsons


one-third rule

113

Table3.16. Example 3.7 Double integration by Gaussian


quadrature

115

Table3.17. Example 3.19 Plate bending

143

Table3.18. Example 3.19 Plate bending

143

Table3.19. Example 3.19 Plate bending

144

Table3.20. Example 3.19 Plate bending

145

Table4.1. Example4.13 Moment-distribution

173

Table4.2. Example4.14 Moment-distribution

174

Table4.3. Example4.20 Global joint stiffness

199

Table4.4. Example4.20 Global joint stiffness

200

Table 4.5. Example 4.20 Global joint stiffness

200

Table4.6. Example4.20 Global joint stiffness

202

Table4.7. Example4.20 Global joint stiffness

202

Table4.8. Example4.20 Global joint stiffness

203

Table4.9. Example4.20 Global joint stiffness

203

Table4.10. Example4.20 Global joint stiffness

203

Table5.1. Release codesX-Z system

214

Table5.2. Release codesX-Y system

217

Table5.3. Example5.3 M
 ember stiffness, member 1

220

Table5.4. Example5.3 Member stiffness, member 4

221

Table5.5. Example5.3 Member stiffness, member 5

222

Table5.6. Example5.3 Member stiffness, member 3

223

Table5.7. Example5.3 Member stiffness, member 2

224

Table5.8. Example5.3 Member stiffness

225

Table5.9. Example5.3 Member stiffness

226

Table 5.10 
Example 5.3 Member stiffness

227

Table 5.11. Example 5.3 Member stiffness

228

Table 5.12. Example 5.3 Member stiffness

228

Table 5.13. Example 5.3 Member stiffness

228

Table 5.14. Example 5.3 Member stiffness

229

Table 5.15. Example 5.3 Member stiffness

229

Table 5.16. Example 5.13 Shear area

251

Acknowledgments
Our sincere thanks go to Associate Professor Christopher M. Papadopoulos,
PhD, Department of Engineering Science and Materials University of
Puerto Rico, Mayagez.
We would also like to thank the ARCH 6243 Structures: AnalysisII, Spring 2014 class: Kendall Belcher, Conner Bowen, Harishma
Donthineni, Gaurang Malviya, Alejandro Marco Perea, Michael
Nachreiner, Sai Sankurubhuktha, Timothy Smith, Nuttapong Tanasap,
Ignatius Vasant, and Lawrence Wilson.
A special thanks to Nicholas Prather for his assistance with figures.

xix

CHAPTER 1

Roots of Algebraic
and Transcendental
Equations
In structural engineering, it is important to have a basic knowledge of how
computers and calculators solve equations for unknowns. Some equations
are solved simply by algebra while higher order equations will require
other methods to solve for the unknowns. In this chapter, methods of finding roots to various equations are explored. The roots of an equation are
defined as values of x where the solution of an equation is true. The most
common roots are where the value of the function is zero. This would
indicate where a function crosses an axis. Roots are sometimes complex
roots where they contain both a real number and an imaginary unit.

1.1EQUATIONS
Equations are generally grouped into two main categories, algebraic equations and transcendental equations. The first type, an algebraic equation,
is defined as an equation that involves only powers of x. The powers of x
can be any real number whether positive or negative. The following are
examples of algebraic equations:
8 x3 3x 2 + 5 x 6 = 0
1
+2 x =0
x
x1.25 3p = 0
The second type is transcendental equations. These are non-algebraic
equations or functions that transcend, or cannot be expressed in terms of

2NUMERICAL STRUCTURAL ANALYSIS

algebra. Examples of such are exponential functions, trigonometric functions, and the inverses of each. The following are examples of transcendental equations:
cos ( x ) + sin ( x ) = 0
e x + 15 = 0
Transcendental functions may have an infinite number of roots or may not
have any roots at all. For example, the function sin(x) =0 has an infinite
number of roots x = kx and k=0, 1, 2.
The solution of algebraic or transcendental equations is rarely carried
out from the beginning to end by one method. The roots of the equation
can generally be determined by one method with some small accuracy,
and then made more accurate by other methods. For the intent and purpose of this text, only a handful of the available methods are discussed.
These methods include: Descartes Rule, Synthetic Division, Incremental
Search, Refined Incremental Search, Bisection, False Position, Secant,
NewtonRaphson, Newtons Second Order, Graeffes Root Squaring, and
Bairstows methods. Some of these methods are used to solve specific
types of equations, while others can be used for both equation types.

1.2POLYNOMIALS
A polynomial is defined as an algebraic equation involving only positive
integer (whole number) powers of x. Polynomials are generally expressed
in the following form:
a0 x n + a1 x n 1 + a2 x n 2 + a3 x n 3 +  + an 1 x1 + an = 0

In most cases, the polynomial form is revised by dividing the entire equation by the coefficient of the highest power of a, a0, resulting in the following form:
x n + a1 x n 1 + a2 x n 2 + a3 x n 3 +  + an 1 x1 + an = 0
For these polynomials, the following apply:

The order or degree of the polynomial is equal to the highest power


of x and the number of roots is directly equal to the degree or n,
where an is not equal to 0. For example, a sixth degree polynomial,
or a polynomial with n = 6 has six roots.
The value of n must be a non-negative integer. In other words, it
must be a whole number that is equal to zero or a positive integer.

Roots of Algebraic and Transcendental Equations 3

The coefficients (a0, a1, a2 , an-1, an) are real numbers.


There will be at least one real root if n is an odd integer.
It is possible that equal roots exist.
When complex roots exist, they occur in conjugate pairs. For
example:
x = u vi = u v 1

1.3 DESCARTES RULE


Descartes rule is a method of determining the maximum number of positive and negative real roots of a polynomial. This method was published
by Ren Descartes in 1637 in his work La Gomtrie (Descartes 1637).
This rule states that the number of positive real roots is equal to the number of sign changes of the coefficients or is less than this number by an
even integer. For positive roots, start with the sign of the coefficient of
the lowest (or highest) power and count the number of sign changes from
the lowest to the highest power (ignore powers that do not appear). The
number of sign changes proves to be the number of positive roots. Using
x=1 in evaluating f(x) =0 is the easiest way to look at the coefficients.
For negative roots, begin by transforming the polynomial to f(x)=0.
The signs of all the odd powers are reversed while the even powers remain
unchanged. Once again, the sign changes can be counted from either the
highest to lowest power, or vice versa. The number of negative real roots
is equal to the number of sign changes of the coefficients, or less than by
an even integer. Using x=1 in evaluating f(x)=0 is the easiest way to
look at the coefficients.
When considering either positive or negative roots, the statement
less than by an even integer is included. This statement accounts for
complex conjugate pairs that could exist. Complex conjugates change the
sign of the imaginary part of the complex number. Descartes rule is valid
as long as there are no zero coefficients. If zero coefficients exist, they are
ignored in the count. Also, one could find a root and divide it out to form a
new polynomial of degree n 1 and apply Descartes rule again.
Example 1.1Descartes rule
Find the possible number of positive, negative, and complex roots for the
following polynomial:
x 3 6 x 2 + 11x 6 = 0

4NUMERICAL STRUCTURAL ANALYSIS

Find possible positive roots for f(x)=0:


x 3 6 x 2 + 11x 6 = 0
1

= 3 sign changes

Since there are three sign changes, there is a maximum of three positive
roots. Three positive real roots exist or one positive real root plus two
imaginary roots.
Find possible negative roots by rewriting the function for f(x)=0:

( x)3 6 ( x)2 + 11( x) 6 = x3 6 x 2 11x 6 = 0


x 3 6 x 2 11x 6 = 0
0

0 sign changes

Notice the signs of all the odd powers reverse while the signs of the even
powers remain unchanged. Count the number of sign changes, n. This
number is the maximum possible negative roots. Since there is no sign
change, zero negative roots exist.
Possible complex roots:
Complex roots appear in conjugate pairs. Therefore, either zero or two
complex roots exist. In this example the roots are x=1, 2, 3.
Example 1.2Descartes rule
Find the possible number of positive, negative, and complex roots for the
following polynomial:
x3 7 x 2 + 6 = 0
Find possible positive roots for f(x)=0:
x3 7 x 2 + 6 = 0
1

2 =

2 sign changes

Since there are two sign changes, there is a maximum of two positive
roots. Two or zero positive real roots exist.

Roots of Algebraic and Transcendental Equations 5

Find possible negative roots by rewriting the function for f(x) = 0:

( x )3 7 ( x ) 2 + 6 = x 3 7 x 2 + 6 = 0
x3 7 x 2 + 6 = 0
0

1 =

1 sign change

Again, the signs of all the odd powers reverse while the signs of the even
powers remain unchanged. Count the number of sign changes, n. This
number is the maximum possible negative roots. Since there is one sign
change, one negative root exists.
Possible complex roots:
Complex roots appear in conjugate pairs. Therefore, either zero or two
complex roots exist. In this example, the roots are x=1, 2, 3.
Example 1.3 Descartes rule
Find the possible number of positive, negative, and complex roots for the
following polynomial:
x3 3x 2 + 4 x 6 = 0
Find possible positive roots for f(x)=0:
x3 3x 2 + 4 x 6 = 0
1

3 = 3 sign changes

Since there are three sign changes, there is a maximum of three positive
roots. Three or one positive real roots exist.
Find possible negative roots by rewriting the function for f(x)=0:

( x )3 3 ( x ) 2 + 4 ( x ) 6 = x 3 3 x 2 4 x 6 = 0
x3 3x 2 4 x 6 = 0
0

0 =

0 sign changes

Count the number of sign changes. Since there is no sign change, zero
negative roots exist.

6NUMERICAL STRUCTURAL ANALYSIS

Possible complex roots:


Complex roots appear in conjugate pairs. Therefore, either zero or two
complex roots exist. In this example the roots are x=1, 1+i, 1i. It
should be noted that the existence of complex conjugate pairs cannot be
readily known. Examples 1.1 and 1.3 had the same number of sign change
count, but the latter had a complex pair of roots.
Example 1.4 Descartes rule
Find the possible number of positive, negative, and complex roots for the
following polynomial:
x3 x 2 + 2 x = 0
Find possible positive roots for f (x)=0:
x3 x 2 + 2 x = 0
1

2 sign changes

Since there are two sign changes, there is a maximum of two positive
roots. Two or zero positive real roots exist.
Find possible negative roots by rewriting the function for f(x) = 0:

( x )3 ( x ) 2 + 2 ( x ) = x 3 x 2 2 x = 0
x3 x 2 2 x = 0
0

0 sign changes

Count the number of sign changes, n. Since there is no sign change, zero
negative roots exist.
Possible complex roots:
Complex roots appear in conjugate pairs. Therefore, either zero or two
complex roots exist. In this example, the roots are x = 0, 1+i, 1i.
The existence of zero as a root could have been discovered by noticing
that there was not a constant term in the equation. Therefore, dividing the
equation by x yields the same as x=0.

Roots of Algebraic and Transcendental Equations 7

1.4 SYNTHETIC DIVISION


Synthetic division is taught in most algebra courses. The main outcome is
to divide a polynomial by a value, r. This is in fact the division of a polynomial, f(x)=0 by the linear equation xr. The general polynomial can
be divided by xr as follows:
f ( x ) = a0 x n + a1 x n 1 + a2 x n 2 + a3 x n 3 +  + an 1 x1 + an = 0
Table 1.1. Synthetic division
r

a0

a1

a2

an-1

an

rb1

rb2

rbn-1

rbn

b1

b2

b3

bn

The results, b, are the sum of the rows above (i.e., b1 = a0 + 0 or


bn=an1+rbn1). If r is a root, then the remainder, R, will be zero. If r is not
a root, then the remainder, R, is the value of the polynomial for f(x) at x=r.
Furthermore, after the first division of a polynomial, divide again to
find the value of the first derivative equal to the remainder times one factorial, R*1!. After the second division of a polynomial, divide again to
find the value of the second derivative equal to the remainder times two
factorial, R*2!. Continuing this process, and after the third division of a
polynomial, divide again to find the value of the third derivative equal to
the remainder times three factorial, R*3!. Basically, two synthetic divisions yield the first derivative, three synthetic divisions yield the second
derivative, four synthetic divisions yield the third derivative, and so on.
Example 1.5Synthetic division
Find f(1) or divide the following polynomial by x1.
x 3 6 x 2 + 11x 6 = 0
Set up the equation as shown below by writing the divisor, r, and coefficient, a, in the first row.
Table 1.2. Example 1.5 Synthetic division
1

-6

11

-6

8NUMERICAL STRUCTURAL ANALYSIS

Add the columns by starting at the left. Multiply each result by r=1 and
add this to the next column.
Table 1.3. Example 1.5 Synthetic division
1

1
0
1

-6

11

-6

-5

-5

6
0

Since the remainder, R, is zero, f(r)=0 and r=1 is a root. The polynomial can now be written as a linear equation, xr or x1, and
the resulting reduced polynomial with coefficient of the resultants as
follows:

x 3 6 x 2 + 11x 6 = ( x 1) x 2 5 x + 6 = 0
Use the quadratic equation to reduce the remaining polynomial as follows:
b b 2 4ac ( 5) ( 5) 4 (1) 6 5 1
x=
=
=
= 2, 3
2a
2 (1)
2
2

If the remaining polynomial was divided by x2, r =2 is a root as


follows:
Table 1.4. Example 1.5 Synthetic division
2

1
0
1

-5

-6

-3

Since the remainder, R, is zero, f(2)=0 and r=2 is a root. The resulting
polynomial is x3, thus x=r is the third root. This can also be shown by
repeating division with x3.
Table 1.5. Example 1.5 Synthetic division
3

1
0
1

-3

3
0

Roots of Algebraic and Transcendental Equations 9

Since the remainder, R, is zero, f(3)=0 and r=3 is a root. The polynomial
is now written as:

x 3 6 x 2 + 11x 6 = ( x 1) x 2 5 x + 6 = ( x 1) ( x 2) ( x 3) = 0
The roots are x=1, 2, 3.
Example 1.6Synthetic division
Find f(1), f (1), and f (1) or perform three divisions of the following
polynomial by x +1:
x 3 6 x 2 + 11x 6 = 0
Set up the equation as shown in the following by writing the divisor, r, and
coefficient, a, in the first row.
Table 1.6. Example 1.6 Synthetic division
1

-1

-6

11

-6

Add the columns by starting at the left. Multiply each result by r=1 and
add this to the next column.
Table 1.7. Example 1.6 Synthetic division
-1

1
0
1

11
7
18

-6
-1
-7

-6
-18
-24

Since the remainder, R, is 24, f(1)=24, the polynomial evaluated at 1


is 24. Performing a check as follows:
f ( 1) = ( 1) 6 ( 1) + 11( 1) 6 = 24
3

Now divide the remaining polynomial again by 1 to find f (1).

10NUMERICAL STRUCTURAL ANALYSIS

Table 1.8. Example 1.6 Synthetic division


-1
-1

1
0
1
0
1

-6
-1
-7
-1
-8

11
7
18
8
26

-6
-18
-24

Since the remainder, R, is 26, f (1)=R*1!=26. The first derivative of the


polynomial evaluated at 1 is 26. Performing a check as follows:

f ( x ) = 3 x 2 12 x + 11 = 0
f ( 1) = 3( 1) 12 ( 1) + 11 = 26
2

Divide the remaining polynomial again by 1 to find f(1).


Table 1.9. Example 1.6 Synthetic division
-1
-1
-1

-6

11

-6

0
1
0
1
0
1

-1

7
18
8
26

-18

-7
-1
-8

-24

-1
-9

Since the remainder, R, is 9, f (1)=R*2!= 9(1)( 2)= 18 and the


second derivative of the polynomial evaluated at 1 is 18. Performing a
check as follows:
f ( x ) = 6 x 12 = 0

f ( 1) = 6 ( 1) 12 = 18

1.5INCREMENTAL SEARCH METHOD


The incremental search method is a simple and quick way to find the
approximate location of real roots to algebraic and transcendental

Roots of Algebraic and Transcendental Equations 11

equations. A search is performed over a given range of values for x usually


denoted as minimum and maximum values, xmin and xmax. An increment on
x of x is used to determine the successive values of f(x). Each consecutive
pair of functions of x are compared and when their signs are different a
root of x has been bounded by the two values of x. Written in algorithmic
form, a sign change occurs between xi and xi+1 if f(xi)f(xi+1) 0. The sign
change generally indicates a root has been passed but could also indicate
a discontinuity in the function. This process is illustrated graphically in
Figure 1.1.
Example 1.7 Incremental search method
Determine the first approximate root of the following function starting at
xmin=0 and using an increment x=0.25:

x 3 8.4 x 2 + 20.16 x 13.824 = 0


Table 1.10. Example 1.7 Incremental search method
x
f(x)

0.25

0.5

0.75

-13.8240

-9.2934

-5.7190

-3.0071

-1.0640

1.25
0.2041

f(x)

f(x)
f(xi)
xi

xi+1
f(xi+1)

Figure 1.1. Incremental search method.

(x)

12NUMERICAL STRUCTURAL ANALYSIS

Since the sign of f(x) changed between x=1 and x=1.25, it is assumed
that a root was passed between those values. The actual root occurs at
x=1.2. In this example, five digits of precision were used, but in most
cases it is a good rule to carry one more digit in the calculations than in the
desired accuracy of the answer.
Once the roots have been bounded by the incremental search method,
other methods can be utilized in finding more accurate roots: The following sections will cover the refined incremental search, bisection, false
position, secant, NewtonRaphson, and Newtons second order methods to determine more accurate roots of algebraic and transcendental
equations.

1.6 REFINED INCREMENTAL SEARCH METHOD


Closer approximations of the root may be obtained by the refined incremental search method. This method is a variation of the incremental
search method. Once a root has been bounded by a search, the last value
of x preceding the sign change is used to perform another search using
a smaller increment such as x/10 as shown in Figure 1.2 until the sign
changes again.
This process can be repeated with smaller increments of x until the
desired accuracy of the root is obtained. Usually the accuracy on the

f(x)

x/10
f(x)
f(xi)
xi

f(xi+1)
xi+1

Figure 1.2. Refined incremental search method.

(x)

Roots of Algebraic and Transcendental Equations 13

function of x is represented by epsilon, e, where |f(x)|<e. Care must


be taken in the selection of the starting point and the increment so that
a root is not missed. This could happen if two roots occur within an
increment and the sign of the function does not change at the successive
values of x.
Example 1.8 Refined incremental search method
Refine the search of the function from Example 1.7 between 1.0 and 1.25
using an increment of x=0.25/10 or 1/10th the original increment.

x 3 8.4 x 2 + 20.16 x 13.824 = 0


Table 1.11. Example 1.8 Refined incremental search method
x
1
1.025
1.05
1.075
f(x)
-1.0640
-0.9084
-0.7594
-0.6170
x
f(x)

1.1
-0.4810

1.125
-0.3514

1.15
-0.2281

1.175
-0.1110

1.2
0.0000

Since the sign of f(x) changed between x=1.175 and x=1.2, it is assumed
that a root was passed between those values. The actual root occurs at
x=1.2.

1.7BISECTION METHOD
After a sign change has occurred in a search method, another way to rapidly converge (become closer and closer to the same number) on a root is
the bisection method, also known as the half-interval method or the Bolzano method developed in 1817 by Bernard Bolzano. This method takes
the bounded increment between two points xi and xi+1 where f(xi)f(xi+1)0
and divides it in two equal halves or bisects the increment. The two subintervals have the first interval from xi to xi+ and the second interval from
xi+ to xi+1 as seen in Figure 1.3.
Next, the subinterval containing the root can be found by the following algorithm:
f(xi)f(xi+ ) < 0, first interval contains the root
f(xi)f(xi+ ) > 0, second interval contains the root
f(xi)f(xi+ ) = 0, xi+ is the root

14NUMERICAL STRUCTURAL ANALYSIS


f(x)

f(x)
f(xi)
xi

f(xi+1/2)
xi+1/2

xi+1

(x)

f(xi+1)
x

Figure 1.3. Bisection method.

The process is continued by bisecting the subinterval containing the root


and repeating the procedure until the desired accuracy is achieved. After
n bisections, the size of the original interval has been reduced by a factor
of 2n.
Example 1.9Bisection method
Refine the search of the function from Example 1.7 between 1.0 and 1.25
using the bisection method to increase the accuracy. Use e=0.01 that is
|f (x)|<e:

x 3 8.4 x 2 + 20.16 x 13.824 = 0


Begin by solving the equation for 1 and 1.25, which was done in
Example 1.7. The sign changes, so a root lies between the two. We also
know from the refined incremental search method the root should fall
between 1.175 and 1.2. Next, bisect the increment between 1 and 1.25,
which is a value of 1.125. Evaluate the function at that point and compare the two subintervals for the sign changes. Also, check to see if
the desired accuracy on f (x) is achieved. This occurs between 1.125
and 1.25, so that interval is subdivided again at 1.1875. Continue the
bisections until the desired accuracy is achieved. Note in Table 1.12
this occurs at x=1.1992 where f(x)=0.0034. This is the last bisection
between 1.1953 and 1.2031.

Roots of Algebraic and Transcendental Equations 15

Table 1.12. Example 1.9 Bisection method


1

1.25

1.125

1.1875

1&2

2&3

-1.0640

0.2041

-0.3514

-0.0548

x
interval
f(x)
5

1.2188

1.2031

1.1953

1.1992

interval

2&4

4&5

4&6

6&7

f(x)

0.0793

0.0135

-0.0204

-0.0034

1.8METHOD OF FALSE POSITION OR LINEAR


INTERPOLATION
Although the bisection method can be used to reach convergence, other
methods such as false position provide the same accuracy more rapidly.
The process is similar to the bisection method in that between xi and xi+1
where f(xi)f(xi+1) 0 a root exists. Refer to xi and xi+1 as x1 and x2, respectively. A straight line connecting x1 and x2 intersects the x-axis at a new
value, say x3, which is closer to the root than either x1 or x2. Thus, by similar triangles, the value of x3 can be found.
f ( x2 ) f ( x1 )
x2 x1

0 f ( x1 )
x3 x1

x3 = x1 f ( x1 )

x2 x1
f ( x2 ) f ( x1 )

This equation can also be rewritten as follows:


x3 =

x1 f ( x2 ) x2 f ( x1 )
f ( x2 ) f ( x1 )

The relationship between x1, x2, and x3 can be seen in Figure 1.4.
f(x1)f(x3) < 0, first interval contains the root
f(x1)f(x3) > 0, second interval contains the root
f(x1)f(x3) = 0, x3 is the root

(1.1)

16NUMERICAL STRUCTURAL ANALYSIS

f(x)

f(x1)

x1

x3

x2

(x)
f(x2)

Figure 1.4. Method of false position or linear interpolation.

If the first interval contains the root, the values for the next cycle for x1 and
x2 and the corresponding functions f(x1) and f(x2) are as follows:
x1 and f(x1) remain unchanged
x2= x3
f(x2)= f(x3)
If the second interval contains the root, then the values are used for x1 and
x2 and the corresponding functions f(x1) and f(x2) are as follows:
x2 and f(x2) remain unchanged
x1=x3
f(x1)=f(x3)
The process is continued until the desired accuracy is obtained.
Example 1.10 Method of false position
Refine the search of the function from Example 1.7 between 1.0 and 1.25
using the false position method to increase the accuracy of the approximate root. For the accuracy test use =0.01 that is |f(x)| < :

x 3 8.4 x 2 + 20.16 x 13.824 = 0

Roots of Algebraic and Transcendental Equations 17

Using Equation 1.1 to solve for a closer point between x1=1 and x2=1.25.
x3 =

x1 f ( x2 ) x2 f ( x1 )
f ( x2 ) f ( x1 )

1( 0.2041) 1.25 ( 1.064)


= 1.2098
0.2041 ( 1.064)

Repeat this process until the desired accuracy of the root is achieved as
shown in Table 1.13.
Table 1.13. Example 1.10 Method of false position
x

1.25

1.2098

1.2018

1&2

1&3

0.0417

0.0080

interval
f(x)

-1.064

0.2041

1.9 SECANT METHOD


The secant method is similar to the false position method except that the
two most recent values of x (x2 and x3) and their corresponding function
values [f(x2) and f(x3)] are used to obtain a new approximation to the root
instead of checking values that bound the root. This eliminates the need to
check which subinterval contains the root. The variable renaming process
for iteration is as follows:
x1= x2andx2= x3
f(x1)= f(x2)andf(x2)= f(x3)
In some instances interpolation occurs, this is when the new value is
between the previous two values. In others, extrapolation occurs, meaning
the new value is not between the previous two values. Interpolation was
shown in Figure 1.4 and extrapolation is shown in Figure 1.5.
Example 1.11 Secant method
Refine the search of the function from Example 1.7 between 1.0 and 1.25
using the secant method to increase the accuracy of the approximate root.
For the accuracy test use e=0.01.

x 3 8.4 x 2 + 20.16 x 13.824 = 0

18NUMERICAL STRUCTURAL ANALYSIS


f(x)

f(x3)
x3
x1

x2

(x)

f(x1)

f(x2)

Figure 1.5. Secant method.

The process of finding the new value is the same as linear interpolation
using Equation 1.1 to solve for a closer point between x1=1 and x2=1.25.
x3 =

x1 f ( x2 ) x2 f ( x1 )
f ( x2 ) f ( x1 )

1( 0.2041) 1.25 ( 1.064)


= 1.2098
0.2041 ( 1.064)

The reassignment of the values simply uses the last two values and their
corresponding functions as shown in Table 1.14.
Table 1.14. Example 1.11 Secant method
x
f(x)

1.25

1.2098

1.1994

1&2

2&3

-1.0640

0.2041

0.0417

-0.0025

This happens to be similar to the false position Example 1.10 as only interpolations occur, but with different sub-intervals.

1.10NEWTONRAPHSON METHOD OR
NEWTONS TANGENT
The NewtonRaphson method uses more information about the function
to speed up convergence. It was originally developed by Issac Newton in

Roots of Algebraic and Transcendental Equations 19


f(x)

f(x)

f(xn)

xn

xn+1

(x)

f(xn)

Figure 1.6. NewtonRaphson method or Newtons tangent.


1669 (Newton 1669). Once an approximate root xn has been found, not
only is the function, f(xn), used, but the slope of the function at that point,
f(xn), is also incorporated to converge to the root more rapidly. The slope
of the function is found from the first derivative of the function evaluated
at a point. This only requires the use of one value to be known. The slope
intersects the x-axis at a value xn+1 as shown in Figure 1.6 and the relationship is given in Equation 1.2.
f ( xn ) =

f ( xn )
f ( xn )
xn+1 = xn
xn xn+1
f ( xn )

(1.2)

Repeat the process using a new value until convergence occurs. Convergence may not occur in the following two cases:

f(xn), (curvature) changes sign near a root, shown in Figure 1.7.


Initial approximation is not sufficiently close to the true root and
the slope at that point has a small value, shown in Figure 1.8.

Example 1.12NewtonRaphson method


Refine the search from Example 1.7 with a starting value of 1.25 using
the NewtonRaphson method to increase the accuracy of the approximate
root. For the accuracy test use e=0.01.

x 3 8.4 x 2 + 20.16 x 13.824 = 0

20NUMERICAL STRUCTURAL ANALYSIS


f(x)

xn+3

xn

xn+1

(x)

xn+2

Figure 1.7. NewtonRaphson method or Newtons tangent.


f(x)

xn

xn+2

xn+4

xn+1 xn+3

(x)

Figure 1.8. NewtonRaphson method or Newtons tangent.

The derivative of the function must be obtained to find the slope at any
given value.

f ( x ) = x 3 8.4 x 2 + 20.16 x 13.824


f ( x ) = 3 x 2 16.8 x + 20.16

Beginning with xn=1.25, use Equation 1.2 to determine the next value.

f (1.25) = 1.253 8.4 (1.25) + 20.16 (1.25) 13.824 = 0.2041


2

f (1.25) = 3(1.25) 2 16.8 (1.25) + 20.16 = 3.8475


xn +1 = xn

f ( xn )

f ( xn )

= 1.25

0.2041
= 1.1969
3.8475

Roots of Algebraic and Transcendental Equations 21

Repeat the process until the desired accuracy is obtained in Table 1.15.
Table 1.15. Example 1.12 NewtonRaphson method
x

1.25

1.1969

1.19999

f(x)
f(x)

0.2041

-0.0132

-0.00004

3.8475

4.3493

4.32010

1.11NEWTONS SECOND ORDER METHOD


Newtons second order method is often a preferred method to determine
the value of a root due to its rapid convergence and extremely close
approximation. This method also includes the second derivative of the
function or the curvature to find the approximate root. The equation
f(x)=0 is considered as the target for the root. Figure 1.9 shows the plot
of the actual function.
The Taylor series expansion was discovered by James Gregory and
introduced by Brook Taylor in 1715 (Taylor 1715). The following is a
Taylor series expansion of f (x) about x=xn:
f ( xn +1 ) = f ( xn ) + f ( xn ) ( x ) +

f ( xn ) ( x )
2!

f ( xn ) ( x )
3!

For a means of determining a value of x that will make the Taylor series
expansion go to zero, the first three terms of the right hand side of the
equation are set equal to zero to obtain an approximate value.
f(x)

f(x)

xn

xn+1

Figure 1.9. Newtons second order method.

(x)

22NUMERICAL STRUCTURAL ANALYSIS

f ( xn ) ( x )
f ( xn ) + ( x ) f ( xn ) +
=0
2

The exact value of x cannot be determined from this equation since only
the first three terms of the infinite series were used in the calculation.
However, a close approximation of the root is a result. When using this
equation to calculate x, a quadratic must be solved yielding two possible
roots. In order to avoid this problem, x = -f (xn)/f (xn) from Newtons
tangent may be substituted into the bracketed term only.

f ( xn ) f ( xn )
f ( xn ) + ( x ) f ( xn )
=0
2 f ( xn )

Solving for x we obtain the following:

f ( xn )

x =

f ( xn ) f ( xn )

f ( xn )
2 f ( xn )

Observing Figure 1.9 we see that x = xn+1 - xn. Substituting into the previous equation, Equation 1.3 is obtained as follows:

f ( xn )

xn+1 = xn

f ( xn ) f ( xn )
f ( xn )

2 f ( xn )

(1.3)

If the first derivative is small, the slope is close to zero near the value
and the next approximation may be inaccurate. Therefore, use the second
derivative term as follows:
f ( xn ) = 0

f ( xn ) ( x )
f ( xn ) + ( x )
=0
2

f ( xn ) ( x )2
f ( xn ) = ( x )

f ( xn )
x2 +
=0
f ( xn )
2

Roots of Algebraic and Transcendental Equations 23

Solving by the quadratic equation, two roots are obtained:


( 0) 02 4 (1)
x =

b b 2 4ac
=
2a

x =

2 (1)

f ( xn )
f ( xn )
2
f ( xn )
f ( xn )
2
2
=
2

f ( xn )

f ( xn )
2

With x = xn+1 - xn, Equation 1.4 is as follows:


xn +1 = xn

f ( xn )

(1.4)

f ( xn )
2

This process is a good tool for finding two roots that are near each other.
This will happen when the slope is close to zero near a root. Double roots
occur when the first derivative is zero, triple roots occur when the first
and second derivatives are zero, and so on. These are shown graphically
in Figure1.10.
Example 1.13 Newtons second order method
Refine the search from Example 1.7 with a starting value of 1.25 using the
Newtons second order method to increase the accuracy of the approximate root. For the accuracy test use e = 0.01.
f(x)

f(x)

f(x)

(x)

Single root

(x)

Double root

Figure 1.10. Newtons second order method.

(x)

Triple root

24NUMERICAL STRUCTURAL ANALYSIS

x 3 8.4 x 2 + 20.16 x 13.824 = 0


The first and second derivatives of the function must be obtained to find
the slope and curvature at any given value.

f ( x ) = x 3 8.4 x 2 + 20.16 x 13.824


f ( x ) = 3 x 2 16.8 x + 20.16
f ( x ) = 6 x 16.8

Beginning with xn=1.25, use Equation 1.3 to determine the next value.

f (1.25) = 1.253 8.4 (1.25) + 20.16 (1.25) 13.824 = 0.2041


2

f (1.25) = 3(1.25) 2 16.8 (1.25) + 20.16 = 3.8475


f (1.25) = 6 (1.25) 16.8 = 9.30

xn +1

f
x
( n)
0.2401
= 1.25

= xn

9.3) 0.2041
f ( xn ) f ( xn )
(

f ( xn )
3.8475

2 (3.8475)
2 f ( xn )

= 1.2001

Repeat the process until the desired accuracy is obtained in Table 1.16.
Table 1.16. Example 1.13 Newtons second order method
x

1.25

1.200143

1.2

f(x)

0.2041

0.00062

0.000000

f(x)

3.8475

4.31863

4.320000

f(x)

-9.3000

-9.59914

-9.600000

1.12GRAEFFES ROOT SQUARING METHOD


Graeffes root squaring method is a root-finding method that was among
the most popular methods for finding roots of polynomials in the 19th
and 20th centuries. This method was developed independently by

Roots of Algebraic and Transcendental Equations 25

Germinal Pierre Dandelin in 1826 and Karl Heinrich Grffe in 1837.


The Graeffes root squaring method is especially effective if all roots
are real. The derivation of this method proceeds by multiplying a polynomial f (x) by f (x) using the following polynomial equations in factored form:
f ( x ) = ( x a1 ) ( x a2 ) ( x a3 )( x an )

f ( x ) = ( 1) ( x a1 ) ( x a2 ) ( x a3 )( x an )

)(

)(

) (

f ( x ) f ( x ) = ( 1) x 2 a12 x 2 a22 x 2 a32 x 2 an2


n

For example, use a third degree polynomial with roots x1, x2, and x3 as
follows:

f ( x ) = 0 = x 3 + a1 x 2 + a2 x + a3
A polynomial with roots x1, x2, and x3 follows:

f ( x ) = 0 = x 3 + a1 x 2 a2 x + a3
Multiplying the two equations together yields the following:

f ( x ) f ( x ) = 0 = x 6 + a12 2a2 x 4 + a22 + 2a1a3 x + a32


Letting y = x2 this equation may be written as follows:

0 = y 3 + a12 2a2 y 2 + a22 2a1a3 y + a32


This equation has roots of x12, x22, and x32. If the procedure was
applied again, another polynomial would be derived with roots of x14,
x24, and x34. If computed a third time, they would be of x18, x28, and
x38. The pattern of the roots becomes clear with continuing cycles. The
general process of an nth degree polynomial would be in the following
forms:

0 = y n + a12 2a2 y n 1 + a22 2a1a3 + 2a4 y n 2

+ a32 2a2 a4 + 2a1a5 2a6 y n 3 +  + an2

a2

a1

(c1/c0)1/m

c1
(c2/c1)1/m

c2

-2b0b6

c0

c3
(c3/c2)1/m

(cn-1/cn-2)1/m

cn-1

-2b2b4
+2b1b5

-2b1b3

bn-1
bn-12

+2b0b4

b3
b32

-2b0b2

b2
b22

b1

-2a0a6

b12

-2a2a4
+2a1a5

-2a1a3
+2a0a4

an-1
2

an-1

an-1x

b0

-2a0a2

a
2
3

a3

a3xn-3

b02

a
2

2
0

a2

a1

a0

a2xn-2

a1xn-1

a0xn

Table 1.17. Graeffes root squaring method

(cn/cn-1)1/m

cn

bn2

bn

a
2
n

an

anx0

cycle

26NUMERICAL STRUCTURAL ANALYSIS

Roots of Algebraic and Transcendental Equations 27

This can be written with the coefficients in a vertical format as follows:


a32
a22

a n 1
n 2 2a2 a4 n 3
0 = yn +
y
+

2
a
a
y
+
y

1 3
2a2
+2a1a5
+2a
4

2a6
a42

2a3 a5
+ +2a2 a6 y n 4 +  + an2
2a a
1 7

2
a
+

2
1

A tabular solution may be set up considering the following general


polynomial:
f ( x) = a0 x n + a1 x n 1 + a2 x n 2 + a3 x n 3 +  + an 1 x1 + an = 0
Carefully inspect the coefficient of the polynomial for a pattern. The solution of the original polynomial can take three different forms. These are
shown in Figure 1.11.

Real and distinct roots


Real and equal roots
Complex roots

f(x)

f(x)

f(x)

(x)

Three real roots

(x)

(x)

One real and


one double root

Figure 1.11. Graeffes root squaring method.

One real and


one imaginary root

28NUMERICAL STRUCTURAL ANALYSIS

1.12.1 REAL AND DISTINCT ROOTS


The first possible solution type will occur after many cycles of squaring
the polynomial; the coefficients of the derived polynomial are the squares
of the terms from the preceding cycle. This is known as the regular solution and yields real and distinct roots (not equal). The roots of the polynomial or the derived polynomials can be determined from the factored form
of the polynomials. The Enke roots of a polynomial are the negatives of
the roots of the polynomial. If r is denoted as the Enke root designation,
then x1 = -r1, x2 = -r2 xn = -rn. The third degree polynomial is shown
in factored form:
f ( x ) = 0 = x 3 + a1 x 2 + a2 x + a3

f ( x ) = 0 = ( x x1 ) ( x x2 ) ( x x3 )
If the previous equation is multiplied out, the following is the result:
f ( x ) = 0 = x 3 ( x1 + x2 + x3 ) x 2 + ( x1 x2 + x1 x3 + x2 x3 ) x x1 x2 x3
Therefore, for the polynomial, the original coefficients are as follows:
a1 = ( x1 + x2 + x3 )
a2 = x1 x2 + x1 x3 + x2 x3
a3 = x1 x2 x3
The Enke roots of x1 = r1, x2 = r2, and x3 = r3 are substituted in. The
sign has been lost so the Enke roots are used as the basis (x1 = r1, x2 = r2,
etc.), then the following is true:
a1 = r1 + r2 + r3
a2 = r1r2 + r1r3 + r2 r3
a3 = r1r2 r3
As the cycles (m) continue, the derived polynomial becomes the following:
f ( x ) = 0 = y 3 + b1 y 2 + b2 y + b3

Roots of Algebraic and Transcendental Equations 29

The Enke root relationship is then as follows:


b1 = r1m + r2m + r3m
b2 = r1m r2m + r1m r3m + r2m r3m
b3 = r1m r2m r3m
If we consider only the dominant terms in each expression, the following
occurs:
b1 r1m
b2 r1m r2m
b3 r1m r2m r3m
These become the following:
1

b1 r1m r1 b1 m
b
b
b
b2 r r r m2 2 r2 = 2
b1
r1
b1
m m
1 2

m
2

b
b
b
b3 r r r r m 3 m 3 r3 = 3
b2
r1 r2
b2
m m m
1 2 3

m
3

The general expression for this regular solution is the following:


b
b
r n rn = n
bn 1
bn 1
m
n

The Enke roots only lack the proper sign and either positive or negative
may be correct, so a check is necessary.
Example 1.14
Graeffes root squaring methodreal and distinct
roots.
Find the root of the following polynomial using Graeffes root squaring
method.
f ( x) = 0 = x 4 10 x 3 + 35 x 2 50 x + 24

30NUMERICAL STRUCTURAL ANALYSIS

Table 1.18. Example 1.14 Graeffes root squaring methodreal and


distinct roots
m

x4

a1x3

a2x2

a3x

a4x0

cycle

-10

35

-50

24

100

1225

2500

576

-70

-1000

-1680

48
2

1
1

30

273

820

576
331776

900

74529

672400

-546

-49200

-314496

1152
4

354

26481

357904

331776

1.253E+05

7.012E+08

1.281E+11

1.101E+11

-5.296E+04

-2.534E+08

-1.757E+10

6.636E+05
8

7.235E+04

4.485E+08

1.105E+11

1.101E+11

5.235E+09

2.012E+17

1.222E+22

1.212E+22

-8.970E+08

-1.599E+16

-9.874E+19

4.338E+09

1.852E+17

1.212E+22

1.212E+22

1.882E+19

3.429E+34

1.468E+44

1.468E+44

-3.703E+17

-1.051E+32

-4.487E+39

2.202E+11
16

2.423E+22
32

1.845E+19

3.418E+34

1.468E+44

1.468E+44

3.404E+38

1.168E+69

2.155E+88

2.155E+88

-6.836E+34

-5.417E+63

-1.004E+79

2.936E+44
64

3.403E+38

1.168E+69

2.155E+88

2.155E+88

1.158E+77

1.365E+138

4.646E+176

4.646E+176

-2.337E+69

-1.467E+127

-5.037E+157

4.311E+88
128
r

1.158E+77

1.365E+138

4.646E+176

4.646E+176

Roots of Algebraic and Transcendental Equations 31

Refer to Table 1.17 for the basic procedure for the root squaring. Table1.18
shows the process for this polynomial.
To determine the proper sign of the roots from the Enke roots, a check
is required.
1

77
r1 b1 m = 1.158 (10)

b
r2 = 2
b

b
r3 = 3
b

b
r4 = 4
b
3

1
128

= 4.000

1.365 (10)138
=
77
1.158 (10)

1
128

4.646 (10)176
=
138
1.365 (10)
4.646 (10)176
=
176
4.646 (10)

= 3.000
1
128

= 2.000
1
128

= 1.000

Substituting the Enke roots into the original equations yields x1 = 4.000,
x2= 3.000, x3 = 2.000, and x4 = 1.000.
1.12.2 REAL AND EQUAL ROOTS
After many cycles of squaring the polynomial, the second possible solution type will occur when the coefficients of the derived polynomial are
the squares of the terms in the preceding cycle with the exception of one
term that is the square of the term in the preceding cycle. This indicates
that two of the roots are equal, the one with the squared term and the
next one to the right. Furthermore, if one term is 1/3 the square of the term
in the proceeding cycle, three of the roots are equalthe term with the
1
/3 squared term and the next two to the right. A similar relationship will
occur if four or more roots are equal. The roots (Enke roots) will have a
relationship similar to the following assuming r1 = r2 and considering only
the dominant terms in each expression:
b1 = r1m + r2m + r3m = r1m + r1m b1 2r1m
b2 = r1m r2m + r1m r3m + r2m r3m = r1m r1m b2 r12 m
b3 = r1m r2m r3m = r1m r1m r3m b3 r12 m r3m

32NUMERICAL STRUCTURAL ANALYSIS

These become the following:


b
b1 2r r1 1
2

m
1

b2 r12 m r1 = (b2 )

2m

= r2

b
b3 r r = b r r3 = 3
b
2m m
1
3

m
2 3

After the multiple roots have been passed, the rest of the terms have the
regular solution relationship and will appear as follows:
b
b
r n rn = n
bn 1
bn 1

m
n

If the second term was the square of the term in the previous cycle, then
the solution would appear as follows, assuming r2 = r3 and considering
only the dominant terms in each expression:
b1 = r1m + r2m + r3m = r1m b1 r1m
b2 = r1m r2m + r1m r3m + r2m r3m = r1m r2m + r1m r2m b2 2r1m r2m
b3 = r1m r2m r3m = r1m r2m r2m b3 r1m r22 m
These become the following:
b1 r1m r1 (b1 )

b
b2 2r r = 2b r r2 = 2
2b1
m m
1 2

m 2m
1 2

b3 r r

m
1 2

2m
1 2

=br

b
r2 = 3
b

2m

= r3

= r3

Similar to the previous case where r1 = r2, after the multiple roots have
been passed, the rest of the terms have the regular solution relationship
and will appear as follows:
b
b
r n rn = n
bn 1
bn 1
m
n

Roots of Algebraic and Transcendental Equations 33

Note the pattern of the powers and the relationship for any other variations
of two roots may be found. Derive one more case for a triple root. If the
first term was 1/3 the square of the term in the previous cycle, it would indicate a triple root or r1 = r2 = r3. If we consider only the dominant terms in
each expression, the following relationships occur:
b1 = r1m + r2m + r3m + r4m = r1m + r1m + r1m b1 3r1m
b2 = r1m r2m + r1m r3m + r1m r4m + r2m r3m + r2m r4m + r3m r4m = r1m r1m
+ r1m r1m + r1m r1m b2 3r12 m
b3 = r1m r2m r3m + r1m r2m r4m + r1m r3m r4m + r2m r3m r4m = r1m r1m r1m b3 r13m
These become the following:
b
b1 3r1m r1 1
3

b
b2 3r12 m r1 = 2
2

2m

= r2 = r3

b3 r13m = b2 r3m r1 = (b3 )

3m

= r2 = r3

After the multiple roots have been passed, the rest of the terms have the
regular solution relationship and will appear as follows:
b
b
r n rn = n
bn 1
bn 1
m
n

Just like with the regular solution for real and distinct roots, the Enke roots
only lack the proper sign and either + or must be checked.
Example 1.15
Graeffes root squaring methodreal and equal
roots
Find the root of the following polynomial using Graeffes root squaring
method.
f ( x) = 0 = x3 + 3x 2 4
Refer to Table 1.17 for the basic procedure for the root squaring. Table 1.19
shows the process for this polynomial.

34NUMERICAL STRUCTURAL ANALYSIS

Table 1.19. Example 1.15 Graeffes root squaring methodreal and


equal roots
m

x3

a1x2

a2x1

a3x0

cycle

1
1

1
1

16
256

1
1

256
65536

1
1

65536
4.295E+09

16

1
1

4.295E+09
1.845E+19

32

1
1

1.845E+19
3.403E+38

64

1
1

3.403E+38
1.158E+77

128

1
1

1.158E+77
1.341E+154

256
r

0
0
24
24
576
-288
288
82944
-16896
66048
4.362E+09
-6.724E+07
4.295E+09
1.845E+19
-1.126E+15
1.845E+19
3.403E+38
-3.169E+29
3.403E+38
1.158E+77
-2.511E+58
1.158E+77
1.341E+154
-1.576E+116
1.341E+154
2

-4
16

3
9
0
9
81
-48
33
1089
-576
513
2.632E+05
-1.321E+05
1.311E+05
1.718E+10
-8.590E+09
8.590E+09
7.379E+19
-3.689E+19
3.689E+19
1.361E+39
-6.806E+38
6.806E+38
4.632E+77
-2.316E+77
2.316E+77
2

1.341E+154
1

Notice that the first term in the table for cycle 9 is the square of the term
in the previous cycles and the following solution applies:
b
r1 1
2
r2 (b2 )

b
r3 3
b2

2.316(10)77
=

2m

= 1.341(10)154

1.341(10)154
=
154
1.341(10)

256

= 2.000

2 ( 256)

256

= 2.000 = r1

= 1.000

Roots of Algebraic and Transcendental Equations 35

Substituting the Enke roots into the original equations yields x1=2.000,
x2=2.000, and x3=1.000.
1.12.3 REAL AND COMPLEX ROOTS
The third possible solution type will occur after many cycles of squaring
the polynomial; the coefficients of the derived polynomial are the squares
of the terms in the preceding cycle, except if one or more terms have a
sign fluctuation, then two of the roots are complexthe one with the sign
fluctuation term and the next one to the right constitutes the complex conjugate pair of roots. The roots (Enke roots) will have a relationship similar
to the following assuming r3 and r4 are the complex conjugate pair of roots
and considering only the dominant terms in each expression:
x3 = Reiq = ( cosq + isinq ) = u + iv

x4 = Re iq = ( cosq isinq ) = u iv

The values i and R for the complex form in polar or Cartesian simple
form are:
i = 1

and R = u 2 + v 2

The form of the coefficients will become the following:

)
= r r + ( r Re q ) + ( r Re q ) + ( r Re q ) + ( r Re q )
= ( r r Re q ) + ( r r Re q ) + ( r R ) + ( r R )
= (r r R )

b1 = r1m + r2m + R m eiq m + e iq m


b2
b3
b4

m m
1 2
1 2

1 2

1 2

2 m

2 m

2 m

These become the following using polar transformations:


b1 = r1m + r2m + 2 R m ( cos mq )

b2 = r1m r2m + 2 R m r1m + r2m + cos mq + R 2 m


b3 = 2 ( r1r2 R ) cos mq + R
m

b4 = r1r2 R

2 m

2m

(r

m
1

+ r2m

+ R2m

36NUMERICAL STRUCTURAL ANALYSIS

If we consider only the dominant terms in each expression, the following


occurs:
b1 = r1m r1 = (b1 )

b
b2 = r r = b r r2 = 2
b
m m
1 2

m
1 2

b3 = 2 ( r1r2 R ) cos mq
m

b4 = r1r2 R 2

Dividing the second and fourth equations:


b
b4
= R2m R 4
b2
b2

2m

Using the fact that R2 = u2 + v2 the following is used to find u and v:


a1 = ( x1 + x2 + x3 + x4 )

a1 = ( x1 + x2 + (u + vi ) + (u vi ))

a1 = ( x1 + x2 + 2u )

Use b1 and b2 to find r1 and r2, then x1 and x2. Use b4 and b2 to find R then
use a1 to find u and R to find v. The Enke roots, once again, only lack the
proper sign and either + or may be correct.
Example 1.16Graeffes root squaring methodreal and
complex roots
Find the root of the following polynomial using Graeffes root squaring
method.
f ( x) = 0 = x 4 + x 3 6 x 2 14 x 12
Refer to Table 1.17 for the basic procedure for root squaring. Table 1.20
shows the process for this polynomial.
The third term has a sign fluctuation thus the previously derived relationships apply and the following is the solution:

Roots of Algebraic and Transcendental Equations 37

Table 1.20. Example 1.16 Graeffes root squaring methodreal and


complex roots
m

x4

a1x3

a2x2

a3x

a3x0

cycle

-6

-14

-12

36

196

144

12

28

-144

-24
2

13

40

52

144

169

1600

2704

20736

-80

-1352

-11520

288
4

1
1

89

536

-8816

20736
429981696

7921

287296

77721856

-1072

1569248

-22228992

6.849E+03

1.898E+06

5.549E+07

4.300E+08

4.691E+07

3.602E+12

3.079E+15

1.849E+17

41472
8

-3.796E+06 -7.601E+11 -1.632E+15


8.600E+08
16

4.311E+07

2.843E+12

1.447E+15

1.849E+17

1.859E+15

8.084E+24

2.094E+30

3.418E+34

-5.686E+12 -1.248E+23 -1.051E+30


3.698E+17
32

1.853E+15

7.959E+24

1.043E+30

3.418E+34

3.434E+30

6.334E+49

1.088E+60

1.168E+69

-1.592E+25 -3.866E+45 -5.441E+59


6.836E+34
64

1
1

3.434E+30

6.334E+49

5.441E+59

1.168E+69

1.179E+61

4.012E+99

2.960E+119

1.365E+138

-1.267E+50 -3.736E+90 -1.480E+119


2.337E+69
128
r

1.179E+61

4.012E+99

1.480E+119

1.365E+138

1.421892602

1.40657599

r1 = (b1 )

b
r2 = 2
b
1

= 1.179 (10)
m

61

1
128

4.012 (10)99
=
61
1.179 (10)

= 3.000
1
128

= 2.000

38NUMERICAL STRUCTURAL ANALYSIS

Substituting the Enke roots into the original equations yields x1 = 3.000
and x2 = -2.000.
b
R = 4
b
2

2m

1.3652 (10)138
=
99
4.012 (10)

256

= 1.414 = 2

Using the fact that R2 = u2 + v2, the following is used to find u and v:
1 = a1 = ( x1 + x2 + 2u ) = (3 2 + 2u ) u = 1
R = u 2 + v2 v = R2 u 2 = 2 1 = 1
Thus results are x3 = -1 + i and x4 = -1 - i.

1.13BAIRSTOWS METHOD
Bairstows method was first published by Leonard Bairstow in 1920
(Bairstow 1920). If we divided a polynomial of nth degree by a quadratic
equation, the result will be a polynomial of n2 degree plus some remainder. This remainder can be used to give a closer approximation of the root
quadratic equation. When the remainder is zero, the quadratic is a root
equation. Bairstows method involves using the remainders from double
synthetic division to approximate the error in an assumed quadratic root
equation of a polynomial. The derivation is omitted from this text, but
may be found in Applied Numerical Methods for Digital Computations,
by James, Smith and Wolford (1977). Look at the process of factoring a
polynomial into a quadratic equation times a polynomial of two degrees
less than the original polynomial as follows:

(x

x n + a1 x n 1 + a2 x n 2 + a3 x n 3 +  + an 1 x1 + an = 0
2

)(

+ ux + v x n 2 + b1 x n 3 + b2 x n 4 +  + bn 3 x1 + bn 2 + remainder = 0

The derived polynomial follows with the terms in the brackets being the
remainder:

(x

n2

+ b1 x n 3 + b2 x n 4 +  + bn 3 x1 + bn 2 + [bn 1 + bn ] = 0

Divide the resulting polynomial by the quadratic equation in order to


derive an equation that has something to do with the derivative of the
original equation. This will be a polynomial four degrees less than the

Roots of Algebraic and Transcendental Equations 39

original polynomial. The following is the form of the second polynomial


with the terms in the brackets being the remainder:

(x

n4

+ c1 x n 5 + c2 x n 6 +  + cn 5 x1 + cn 4 + [ cn 3 + cn 2 + cn 1 ] = 0

The solution may be set up in synthetic division form shown in Table 1.21:
Table 1.21. Bairstows method
a0

a1

a2

-ub0

-ub1

-v

-vb0

-u

b0

b1

b2

-uc0

-uc1

-vc0

c0

c1

c2

-u

-v

an-3

an-2

an-1

an

-ubn-4

-ubn-3

-ubn-2

-ubn-1

-vbn-5

-vbn-4

-vbn-3

-vbn-2

bn-3

bn-2

bn-1

bn

-ucn-4

-ucn-3

-ucn-2

-vcn-5

-vcn-4

-vcn-3

cn-3

cn-2

cn-1

Using the preceding values, the approximations for the change in u and v
values denoted u and v are as follows:
bn cn 2
bn 1 cn 3
u =
cn 1 cn 2
cn 2 cn 3

cn 1 bn
cn 2 bn 1
and v =
cn 1 cn 2
cn 2

cn 3

u2 = u + u and v2 = v + v
Continue the process until u and v are equal to zero. The two roots are
as follows by the quadratic equation:

(x

+ ux + v

with x1, 2 =

u u 2 4v
2

Example 1.17 Bairstows method


Find all the roots of the following polynomial using Bairstows method.
f ( x) = 0 = x 5 3 x 4 10 x 3 + 10 x 2 + 44 x + 48
Begin by assuming u = 1.5 and v = 1.5 to perform the synthetic division
shown in Table 1.22.

40NUMERICAL STRUCTURAL ANALYSIS

Table 1.22. Example 1.17 Bairstows method


1

-1.5

-3

-10

10

44

48

-1.5

6.75

7.125

-35.813

-22.969

-1.5

6.75

7.125

-35.813

-4.5

-4.75

23.875

15.313

-10.781

-1.5

-4.125

-43.125

-1.5

-4.125

2.75

28.75

-31.938

-1.5
1

-1.5
-1.5

-6

10.781 28.750
15.313 2.750
( 10.781)( 2.750) ( 28.750)(15.313)
u =
=
31.938 28.750 ( 31.935)( 2.750) ( 28.750) ( 28.750)
28.750 2.750
469.88
914.39
31.938
28.750
v =
31.938
28.750
=

= 0.5139
10.781
15.313
( 31.9938)(15.313) ( 10.781)( 28.750)
=
28.750
( 31.935)( 2.750) ( 28.750) ( 28.750)
2.750

179.08
= 0.1958
914.39
u2 = u + u = 1.5 + 0.5139 = 2.0139
v2 = v + v = 1.5 + 0.1958 = 1.6958

Now, repeat the process using the revised values for u and v shown in
Table 1.23.
Table 1.23. Example 1.17 Bairstows method
-2.0139

-3

-10

10

44

48

-2.014

10.097

3.219

-43.745

-5.972

-1.696

8.503

2.711

-36.837

-5.014

-1.599

21.722

2.965

5.191

-2.014

14.153

-21.868

-23.707

-1.696

11.918

-18.415

10.859

11.772

-39.157

-1.6958
-2.0139

-1.6958
1

-7.028

Roots of Algebraic and Transcendental Equations 41

5.191
2.965
u =
39.157
11.772
=

11.772
10.859
(5.191) (10.859) (11.772)( 2.965)
=
11.772 ( 39.157)(10.859) (11.772)(11.772)
10.859

21.45
= 0.0381
563.77

39.157
11.772
v =
39.157
11.772
=

5.191
2.965
( 39.157)( 2.965) (5.191)(11.772)
=
11.772 ( 39.157)(10.859) (11.772) (11.772)
10.859

177.22
= 0.3145
563.77
u2 = u + u = 2.0139 + 0.0381 = 1.9758
v2 = v + v = 1.6958 + 0.3145 = 2.0102

Now, repeat the process using the revised values for u and v shown in
Table 1.24.
Table 1.24. Example 1.17 Bairstows method
-1.9758

-3

-10

10

44

48

-1.976

9.831

4.305

-48.027

-0.698

-2.010

10.002

4.380

-48.863

-4.976

-2.179

24.308

0.353

-1.561

-1.976

13.735

-18.861

-38.373

-2.010

13.974

-19.189

9.546

19.421

-57.208

-2.0102
-1.9758

-2.0102
1

-6.952

1.561 19.421
0.353 9.546
( 1.561) (9.546) (19.421)(0.353)
u =
=
57.208 19.421 ( 57.208)(9.546) (19.421)(19.421)
19.421 9.546
=

21.74
= 0.0236
923.27

42NUMERICAL STRUCTURAL ANALYSIS

57.208 1.561
19.421 0.353
( 57.208)(0.353) ( 1.561)(19.421)
v =
=
57.208 19.421 ( 57.208)(9.546) (19.421) (19.421)
19.421 9.546
=

10.11
= 0.0110
923.27
u2 = u + u = 1.9758 + 0.0236 = 1.999
v2 = v + v = 2.0102 0.0110 = 1.999

It appears the values are u = 2 and v = 2. Repeat the process using the
revised values for u and v shown in Table 1.25.
Table 1.25. Example 1.17 Bairstows method
-2
-2

-3
-2

10
4
10
24

-10

10
-2
-2

-5

44

48
0
-48
0

-48

4
0

Since the remainders of the first division bn1 and bn are both zero, u =
2 and v = 2 are the coefficients of the root quadratic. Substitute into the
quadratic equation to find the roots.

(x

+ 2 x + 2 with x1, 2 =

2 22 4 ( 2)
2

= 1 1 = 1 i

The first two roots are x1 = 1 + i and x2 = 1 i. The remaining values


are the coefficients of the factored polynomial.

)(

f ( x) = x 2 + 2 x + 2 x 3 5 x 2 2 x + 24

The remaining polynomial may be solved using the same method. This
time begin with u = 0 and v = 0 in Table 1.26.
Table 1.26. Example 1.17 Bairstows method
0

0
0

-5
0

-5
0

-5

-2
0
0
-2
0
0
-2

24
0
0
24

Roots of Algebraic and Transcendental Equations 43

24 5
2 1
( 24)(1) ( 5) ( 2) = 14 = 0.5185
=
u =
2 5 ( 2)(1) ( 5) ( 5) 27
5 1
2 24
5 2 ( 2) ( 2) ( 24)(5) 124
=
v =
=
= 4.5626
2 5 ( 2)(1) ( 5) ( 5) 27
5 1
u2 = u + u = 0 0.0236 = 0.5185
v2 = v + v = 0 4.5626 = 4.5626
Repeat the process using the revised values for u and v shown in
Table1.27.
Table 1.27. Example 1.17 Bairstows method
0.5185

-2

24

-2.324

0.139

4.593

-20.582

-4.481

0.269

3.558

0.519

-2.055

-5
0.519

4.5926
0.5185

4.5926

4.593
1

-3.963

2.807

3.558 3.963
0.269
1
(3.558)(1) ( 3.963) (0.269)
u =
=
2.807 3.963 ( 2.807)(1) ( 3.963) ( 3.963)
1
3.963
=

4.623
= 0.3584
12.900

2.807 3.558
3.963 0.269
( 2.807)(0.269) (3.558) ( 3.963)
v =
=
2.807 3.963
( 2.807)(1) ( 3.963) ( 3.963)
1
3.963
=

14.844
= 1.1516
12.9007

44NUMERICAL STRUCTURAL ANALYSIS

u2 = u + u = 0.5185 0.3584 = 0.8769


v2 = v + v = 4.5926 1.1516 = 5.7742
Repeat the process using the revised values for u and v shown in
Table1.28.
Table 1.28. Example 1.17 Bairstows method
0.8769

-5

-2

24

0.877

-3.616

0.113

5.744

-23.684

-4.123

0.129

0.429

0.877

-2.847

5.7442
0.8769

5.7442

5.744
1

-3.246

3.026

0.429 3.246
0.129
1
(0.429)(1) ( 3.246) (0.129)
u =
=
3.026 3.246 (3.026)(1) ( 3.246) ( 3.246)
1
3.246
=

0.848
= 0.1129
7.511

3.026 0.429
3.246 0.129
(3.026)(0.129) (0.429) ( 3.246)
v =
=
3.026 3.246
(3.026)(1) ( 3.246) ( 3.246)
1
3.246
=

1.783
= 0.2374
7.511

u2 = u + u = 0.8769 0.1129 = 0.9898


v2 = v + v = 5.7442 0.2374 = 5.9816
Repeat the process using the revised values for u and v shown in
Table1.29.

Roots of Algebraic and Transcendental Equations 45

Table 1.29. Example 1.17 Bairstows method


0.9898

-5

-2

24

0.990

-3.969

0.012

5.9816
0.9898

5.982

-23.987

-4.010

0.012

0.025

0.990

-2.990

-3.020

3.004

5.9816

5.982

0.025 3.020
0.012
1
(0.025)(1) ( 3.020) (0.012)
u =
=
3.004 3.020 (3.004)(1) ( 3.020) ( 3.020)
1
3.020
0.061
= 0.0100
6.116
3.004 0.025
3.020 0.012
(3.004)(0.012) (0.025) ( 3.020)
v =
=
3.004 3.020
(3.004)(1) ( 3.020) ( 3.020)
1
3.020
=

0.111
= 0.0181
6.116
u2 = u + u = 0.9898 0.0100 = 0.9998
v2 = v + v = 5.9816 0.0181 = 5.9997

It appears the values are u = 1 and v = 6. Repeat the process using the
revised values for u and v as shown in Table 1.30.
Table 1.30. Example 1.17 Bairstows method
1

-5

-2

-4

-4

6
0

6
1

24
0
-24

Since the remainders of the first division bn-1 and bn are both zero, u=1
and v = 6 are the coefficients of the root quadratic. Substitute them into
the quadratic equation to find the roots.

46NUMERICAL STRUCTURAL ANALYSIS

(x

1x 6 with x1, 2 =

( 1)2 4 ( 6)
2

= 0.5 2.5 = 2, 3

The first two roots are x3 = 2 and x4 = 3. The remaining values are the
coefficients of the factored polynomial.

)(

f ( x) = x 2 + 2 x + 2 x 2 1x 6 ( x 4)
The last root is x5 = 4, which is the value in the remaining polynomial of
degree one x r = x 4. The final factored form of the original quadratic
can be written and the five roots are x1 = 1+i, x2 = 1i, x3 = 2, x4 = 3,
and x5 = 4.

f ( x) = x 2 + 2 x + 2 ( x + 2) ( x 3) ( x 4)

REFERENCES
Bairstow, L. 1920. Applied Aerodynamics. London, England: Longmans, Green
and Co.
Descartes, R. 1637. Discours de la mthode pour bien conduire sa raison, et
chercher la vrit dans les sciences. Lieden, Netherlands.
James, M.L., G.M. Smith, and J.C. Wolford. 1977. Applied Numerical Methods
for Digital Computations. 2nd ed. Boston, Massachusetts: Addison-Wesley
Longman Publishing Co., Inc.
Newton, I. 1669. De analysi per aequationes numero terminorum infinitas,
London, England.
Taylor, B. 1715. Methodus Incrementorum Directa et Inversa. London, England.

CHAPTER 2

Solutions of Simultaneous
Linear Algebraic Equations
Using Matrix Algebra
Matrix algebra is commonly utilized in structural analysis as a method of
solving simultaneous equations. Each motion, translation or rotation, at
each discrete location in a structure is normally the desired variable. This
chapter explores matrix terminology, matrix algebra, and various methods
of linear algebra to determine solutions to simultaneous equations.

2.1 SIMULTANEOUS EQUATIONS


The solutions of simultaneous equations in structural analysis normally
involve hundreds and even thousands of unknown variables. These solutions are generally linear algebraic equations. A typical linear algebraic
equation with n unknowns is as follows, where a is the coefficient, x is the
unknown, and C is the constant. In some cases, x, y, and z are used in lieu
of x1, x2, etc.
a1 x1 + a2 x2 + a3 x3 + + an xn = C
Equation sets can be separated into two categories, homogeneous and
non-homogeneous. Homogeneous equation sets are those in which all the
Cs are zero and all other equation sets are known as non-homogeneous.
A unique solution to a non-homogeneous set exists only if the equations
are independent or non-singular (determinant is non-zero), and a non-trivial
solution set exists to a homogeneous set only if the equations are not independent (determinant is zero). The determinant is further discussed in

48NUMERICAL STRUCTURAL ANALYSIS

Section 2.3. In comparison to a non-trivial solution, a trivial solution is


one where all the unknowns in the equations are equal to zero.
The typical set of n equations with n unknowns is as follows:
a11 x1 + a12 x2 + a13 x3 + + a1n xn = C1
a21 x1 + a22 x2 + a23 x3 + + a2 n xn = C2
a31 x1 + a32 x2 + a33 x3 + + a3n xn = C3


an1 x1 + an 2 x2 + an 3 x3 + + ann xn = Cn
These equations can be written in matrix form, [A][x]=[C] as follows:
a11
a
21
a31


an1

a12
a22
a32

an 2

a13
a23
a33

an 3







a1n x1 C1
a2 n x2 C2
a3n x3 = C3

  
ann xn Cn

2.2MATRICES
A matrix can be defined as a rectangular array of symbols or numerical
quantities arranged in rows and columns. This array is enclosed in brackets and if there are n rows and m columns, the general form of this matrix
is expressed by the following:
a11
a
21
A
=
[ ] a31


an1

a12
a22
a32

an 2

a13
a23
a33

an 3







a1m
a2 m
a3m


anm

A matrix consisting of n rows and m columns is defined as a matrix of


order n m. The relationship between the number of rows and the number
of columns is arbitrary in a general matrix. Many types of matrices exist,
such as row, column, diagonal, square, triangular, identity, and invert.
These are discussed in the following sub-sections.

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 49

2.2.1 ROW AND COLUMN MATRICES


A row matrix is a matrix that reduces to a single row (n = 1).

[ A] = [ a11

a12

a13  a1m ]

Similar to a row matrix, a column matrix is a matrix that reduces to a


single column (m = 1).
a11
a
21
[ A] = a31


an1

2.2.2 SQUARE MATRIX


A matrix in which the number of rows is equal to the number of columns
(n = m) is referred to as a square matrix.
a11
a
21
[ A] = a31


an1

a12
a22
a32

an 2

a13
a23
a33

an 3







a1n
a2 n
a3n


ann

Square matrices are unique because they are the only matrix that has a
reciprocal or invert as described later in this section. Several types of
square matrices exist such as the diagonal matrix, the identity matrix, the
triangular matrix, and the invert matrix.
2.2.3 DIAGONAL MATRIX
A diagonal matrix is defined as a matrix where all elements outside of the
principal diagonal are equal to zero. The diagonal running from the upper
left corner of the array to the lower right corner is considered the principal
diagonal.

50NUMERICAL STRUCTURAL ANALYSIS

a11
0

[ A] = 0


0

0
a22
0

0

0
0
a33

0

 0
 0
 0

 
 ann

2.2.4 IDENTITY MATRIX


An identity matrix is a diagonal matrix where all of the elements along the
principal diagonal are equal to one and is denoted [I].
1
0

[ I ] = 0


0

0 0
1 0
0 1
 
0 0







0
0
0


1

2.2.5 TRIANGULAR MATRIX


When all of the elements on one side of the principal diagonal are zero,
this matrix is a triangular matrix. There are two types of triangular
matrices, upper and lower. An upper triangular matrix, [U], is when all of
the elements below the principal diagonal are zero, and a lower triangular
matrix, [L], occurs when all of the elements above the principal diagonal
are zero.
a11
0

[U ] = 0


0
a11
a
21
[ L ] = a31


an1

a12
a22
0

0
0
a22
a32

an 2

a13
a23
a33

0







0 
0 
a33 
 
an 3 

a1n
a2 n
a3n


ann
0
0
0


ann

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 51

2.2.6 INVERTED MATRIX


Only square matrices where the determinant is not equal to zero (|A| 0)
can have an inverse or reciprocal. These matrices are called non-singular,
which implies that reciprocals of rectangular matrices do exist. Theinverse
of a matrix is defined as follows:

[ I ] = [ A][ A]1
2.2.7 MATRIX MINOR
The matrix minor, [Aij], is found by omitting the ith row and the jth column
of a matrix and writing the remaining terms in a matrix of one size smaller
in rows and columns. It is used in the computation of the determinant. For
example, the minor, [A22] is shown in the following. Note that i = 2 and j = 2:
a11
a
21
[ A22 ] = a31


an1

a12
a22
a32


a13
a23
a33


an 2

an 3

 a1m
a
 a2 m 11
a
 a3m = 31

 
a
 anm n1

a13
a33

an 3

 a1m
 a3m
 

 anm

2.2.8 TRANSPOSED MATRIX


The transposed matrix is found by writing the aij elements of a matrix as
the aji elements of the matrix, [A]T.
a11
a
21
A
=
[ ] a31


an1

[ A]

a11
a
12
= a13


a1m

a12
a22
a32

an 2
a21
a22
a23

a2 m

a13
a23
a33

an 3
a31
a32
a33

a3m

a1m
a2 m
a3m


anm
 an1
 an 2
 an 3

 
 amn







52NUMERICAL STRUCTURAL ANALYSIS

2.3 MATRIX OPERATIONS


2.3.1 MATRIX ADDITION AND SUBTRACTION
Matrices of the same size can easily be added or subtracted. Addition is
achieved by adding the terms with the same row and column position.
Forexample, if matrix [A] and [B] were to be added to obtain matrix [C],
the following equations would be valid:

[ A ] + [ B ] = [C ]
a11 + b11 = c11
a12 + b12 = c12

aij + bij = cij
Matrix subtraction follows the same form as addition where the matrices
are of the same size.

[ A ] [ B ] = [C ]
a11 b11 = c11
a12 b12 = c12

aij bij = cij

Example 2.1 Matrix addition and subtraction


Add matrix [A] and [B] and then subtract matrix [B] from [A].
2 4 6

[ A] = 7 9 3
6 5 1

and

1 2 4

[ B ] = 6 5 2
4 3 0

Addition:
2 4 6 1 2 4 3 6 10
[ A] + [ B ] = 7 9 3 + 6 5 2 = 13 14 5
6 5 1 4 3 0 100 8 1

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 53

Subtraction:
2 4 6 1 2 4 1 2 2
A

B
=
[ ] [ ] 7 9 3 6 5 2 = 1 4 1
6 5 1 4 3 0 2 2 1

2.3.2 SCALAR MULTIPLICATION


Scalar multiplication consists of multiplying a matrix by a scalar.
When this occurs, every entry is multiplied by that scalar as seen in the
following:
a11
a
21
c [ A] = c a31


an1

a12
a22
a32

an 2

a13  a1m ca11


a23  a2 m ca21
a33  a3m = ca31

   
an 3  anm can1

ca12
ca22
ca32

can 2

ca13  ca1m
ca23  ca2 m
ca33  ca3m




can 3  canm

2.3.3 MATRIX MULTIPLICATION


Matrix multiplication proves a little more complicated than scalar
multiplication. In order for two matrices to be multiplied, the number
of columns in the first matrix must equal the number of rows in the
second matrix. The resulting product consists of the same number of
rows as the first matrix and the same number of columns as the second
matrix.

[ A]n m [ B ]m o = [C ]n o
Each term of the product matrix (row i and column j) is obtained by
multiplying each term in row i of the first matrix by the term in row j of
the second matrix and then summing these products.
cij = ai1 b1 j + ai 2 b2 j + + ain bnj

54NUMERICAL STRUCTURAL ANALYSIS

Example 2.2 Matrix multiplication


Multiply matrix [A] and [B] to get [C].
2
3
[ A] =
1

4
2
0

and

1 4 2 3 2 4

2 3 2 4 1 1

[ B] =

The product of a 4 2 matrix multiplied with a 2 6 matrix is a 4 6


matrix.

[ A]4 2 [ B ]2 6 = [C ]4 6
2
3
[ A][ B ] = 1

4
10 20 12 22 8 12
7 18 10 17 8 14
2 1 4 2 3 2 4

=
C
=
[
]
1 4 2 3 2 4
0 2 3 2 4 1 1

4
11 24 14 25 10 16

The first two elements are computed as follows:


c11 = 2 (1) + 4 ( 2) = 2 + 8 = 10

c12 = 2 ( 4) + 4 (3) = 8 + 12 = 20
2.3.4 MATRIX DETERMINANTS
A determinant is only defined for square matrices and can be easily
achieved through expansion by minors of a row or a column when dealing
with small matrices. A row expansion is as follows:
A = a11 A11 a12 A12 + (1)1+ j a1 j A1 j + (1) n +1 a1n A1n
a11
a21

an1

a12
a22

an 2

 a1n

a22
 a2 n
a32
= a11



 anm
an 2

+ + (1)1+ j a1 j

a23  a2 n
a33  a3n
a12



an 3  anm

a21
a31
A1 j + (1) n +1 a1n

an1

a21
a31

an1

a22  a2, n 1
a32  a3, n 1


an 2  an , n 1

a23  a2 n
a33  a3n



an 3  anm

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 55

If |A| is the determinant of the matrix [A], then the following equations are
valid when row k and column k are expanded:
n

A = (1) k + j akj Akj

and

A = (1)i + k aik Aik


i =1

j =1

The basket weave method may be used for a three-by-three determinant


only. Take the sum of the products of the three down-right diagonals minus
the sum of the product of the three up-right diagonals shown as follows:
a11
a21
a31

a12
a22
a32

a13
a11
a23 = a21
a33
a31

a12
a22
a32

a13
a23
a33

a11
a21
a31

a12
a22
a32

a11
a21
a31

a12
a22
a32

a13
a11
a23 = a21
a31
a33

a12
a22
a32

a13
a23
a33

a11
a21
a31

a12
a22
a32

(a11a22 a33 + a12 a23 a31 + a13 a21a32 ) (a13 a22 a31 + a11a23 a32 + a12 a21a33 )
Also, a determinant may be found by the product of the diagonal of any
triangular matrix.
Example 2.3 Matrix determinants
Find the determinant of the following matrix, [A], by expansion of minors
and by the basket weave method.
3 4 6
[ A] = 1 2 3
2 1 2
Expansion of row 1 yields:
A =3

2 3
1 3
1 2
4
+6
1 2
2 2
2 1

A = 3 ( 4 3) 4 ( 2 6) + 6 (1 4)
A = 3 + 16 18 = 1

56NUMERICAL STRUCTURAL ANALYSIS

Basket weave method yields:


3 4 6 3 4 6 3 4
A= 1 2 3 = 1 2 3 1 2
2 1 2 2 1 2 2 1
A = 3 ( 2)( 2) + 4 (3)( 2) + 6 (1)(1) 4 (1)( 2) + 3 (3)(1) + 6 ( 2)( 2)
A = 42 41 = 1

2.4CRAMERS RULE
Two common methods to solve simultaneous equations exist. One is the
elimination of unknowns by elementary row operations and the second
involves the use of determinates. One of the methods involving determinates is known as Cramers rule. This method was published in 1750 by
Gabriel Cramer (1750). The procedure for Cramers rule in the solution to
n linear equations with n unknowns is as follows:
x1 =

A1
A

c1
c2
A1 = c3


a12
a22
a32


a13
a23
a33


cn

an 2

an 3

a11
a21
A3 = a31

an1

a12
a22
a32

an 2

, x2 =







A2
A

, x3 =

A3
A

, , xn =

a1n
a11
a2 n
a21
a3n , A2 = a31


ann
an1

c1
c2
c3

cn

c1  a1n
a11
c2  a2 n
a21
c3  a3n , , An = a31
  

cn  ann
an1

a13
a23
a33

an 3
a12
a22
a32

an 2

An
A





a13
a23
a33

an 3

a1n
a2 n
a3n ,

ann






c1
c2
c3

cn

As you might see, |A1| is the original coefficient matrix, [A], with columnone replaced with the constant column matrix, [c]. The solution to
n simultaneous equations by Cramers rule requires (n1)*(n+1)! multiplications. In other words, the solution of ten simultaneous equations by
determinants would require (9)*(11!) =359,251,200 multiplications.

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 57

Example 2.4 Cramers rule


Find the solution set to the following non-homogeneous linear algebraic
equations using Cramers rule.
2 x1 + 8 x2 + 2 x3 = 14
x1 + 6 x2 x3 = 13
2 x1 x2 + 2 x3 = 5
2 8 2
A = 1 6 1 = 36
2 1 2
14 8 2
A1 = 13 6 1 = 180
5 1 2
2 14 2
A2 = 1 13 1 = 36
2 5 2
2 8 14
A3 = 1 6 13 = 72
2 1 5
x1 =
x2 =
x3 =

A1
A
A2
A
A3
A

180
=5
36

36
=1
36

72
= 2
36

The determinants are shown by row expansion in Table 2.1.

2.5METHOD OF ADJOINTS OR COFACTOR


METHOD
The solution to a set of linear algebraic equations can be achieved by using
the invert of a matrix. The cofactor and adjoint matrices are very helpful
in finding this invert by utilizing determinants.The cofactor matrix is one
where the elements of the matrix are cofactors. Each term in the cofactor

58NUMERICAL STRUCTURAL ANALYSIS

Table 2.1. Example 2.4 Cramers rule


A

2
1

8
6

-1

2
-1
2
-

-1

-1
2

22

32

14
13

8
6

A1

-1

A1

14

-1
2
-180

14

-1
2

-1
2

-42

-36

13

72

14
13

13

-1

13

-1

13

2
-1
2

A2

62

56

A3

2
1
2

8
6
-1

14
13
5

86

13

2
1

-1
2

-86

A3

-36

A2

248

-26

154

A3

-1
2

-1
2

-1
A1

A2

13

-1

-168

-182

14

-1

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 59

of a matrix, Cij, is given in the following equation where |Aij| is the determinant of the minor as defined in Section 2.2.
Cij = ( 1)

i+ j

Aij

Therefore, given the matrix [A], the cofactor matrix is shown as follows.
The matrix of cofactors should not be confused with the constant matrix
of the original linear algebraic equation, although they have the same variable, [C]:
a11
a
21
A
=
[ ] a31


an1

a12

a13
a23
a33

an 3

a22
a32

an 2

[C ] = [ A]cofactor







C11 C12
C
21 C22
= C31 C32



Cn1 Cn 2

a1m
a2 m
a3m


anm
C13
C23
C33

Cn 3







C1m
C2 m
C3m


Cnm

Once the cofactor matrix is known, the invert can be easily calculated:
T
C]
[
=

A
1

The solutions to the simultaneous equations are now found from matrix
multiplication.
[ A][ x] = [C ] Or [ A1 ][C ] = [ x]

Similar to the cofactor method, the method of adjoints, Adj[A], is another


common way to solve for the invert of a matrix. This method is as follows:
A1 =

Adj [ A]
A

60NUMERICAL STRUCTURAL ANALYSIS

The adjoint matrix, Adj[A], is simply the transpose of the cofactor matrix.
This can be expressed in a few ways.
Adj[ A] = [C ]T = [ Acofactor ]T

c11
c
21
Adj [ A] = c31


cn1

c12
c22
c32

cn 2

Or ci j = (1)( i + j ) [ Aj i ]
T

c13  c1m
c11

c
c23  c2 m
12
c33  c3m = c13

  


c1m
cn 3  cnm

c21

c31

c22
c23

c2 m

c32
c33

c3m

 cn1
 cn 2
 cn 3

 
 cmn

It is noted that the subscripts of the adjoint matrix are the reverse of the
cofactor matrix. The main difference is that the transpose is performed
during the operation of taking the adjoint, while in the cofactor method is
done at the end.
Example 2.5 Cofactor method
Find the solution set to the following nonhomogeneous linear algebraic
equations using the cofactor method.
x1 + x2 + x3 + x4 = 10
8 x1 + 4 x2 + 2 x3 + x4 = 26
x1 + x2 x3 + x4 = 2
3 x1 + 2 x2 + 1x3 + 0 x4 = 10

[C ]

A =
1

And Cij = ( 1)

i+ j

Aij

The determinants are shown by row expansion for the 4 4 matrix and by
the basket weave for all the 3 3 matrices in Table 2.2.
The last step in Table 2.2 is to multiply the invert of A, [A]1, times the
constant vector, [C], to get the final solution vector, [x].
Example 2.6 Method of adjoints
Determine the solution to the following set of equations using the adjoint
method. Use the basket weave method for determinants.

A-1

c41=

c31=

c21=

2
-1
1

4
1
2

c11=

c12=

c31 = 1

-0.2500
0.0000
0.7500
0.5000

3
-4
1
6

1
2
-1

0
4
-4
-12

c41 = 6

1
4
1

0.3333
-0.3333
-0.3333
0.3333

-9
4
5
-6

1
1
1

c32=

c42=

-6
-4
-2
12
-0.0833
0.3333
-0.4167
0.1667

c42 = -12

(-1)4+1

c32 = -4

(-1)

c22 = 4
3+1

1
1
0

1
4
2

1
2
1

c21 = -4

c22=

c12 = 0

(-1)

-9

2+1

1
1
0

1+1

(-1)

c11 = 3

1
1
0

1
1
0

1
1
1
0

1
1
2

1
-1
1

2
-1
1

4
1
2

1
2
-1
1

1
4
1
2

1
8
-1
3

Table 2.2. Example 2.5 Cofactor method

CT

1
8
-1

1
8
3

1
1
0

2
-1
1

10
26
2
10

1
1
1

1
1
0

1
1
0

-0.5000
1.0000
0.5000
1.0000

3
0
-9
-6

c43 = -6

1
2
-1

c33 = 5

1
2
1

c23 = 4

1
-1
1

c13 = -9

2
-1
1

8
-1
3
1
-1
3

-6

8
-1
3

-4
4
4
-4

(-1)4+2

(-1)
3+2

(-1)

1+2

2+2

(-1)

1
1
0
8
-1
3

-12

1
-1
3
1
8
3
1
8
-1

1
-4
5
-2

6
-12
-6
12

c44 = 12

c43=

c34 = -2

c33=

c24 = -4

c23=

c14 = -6

c13=

1
2
3
4

1
4
1

1
4
2

1
1
2

4
1
2

8
-1
3

1
1
1

1
1
0

1
1
0

1
1
0

4
1
2

(-1)4+3

(-1)3+3

(-1)2+3

(-1)1+3

1
1
0

c44=

c34=

c24=

c14=

1
8
1

1
8
3

1
-1
3

8
-1
3

1
4
1

1
4
2

1
1
2

4
1
2

8
-1
3

1
2
1

1
2
1

1
-1
1

2
-1
1

4
1
2

(-1)4+4

(-1)3+4

(-1)2+4

(-1)1+4

2
-1
1

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 61

2
-1
0

4
1
0

c11=

A-1

c14=

c13=

c12=

-6

-0.5000
0.5000
1.0000
0.0000

-6
6
12
0

c14 = 6

1
4
1

1
2
0

2
0
-2
0

1
2
-1

c13 = -2

1
4
0

c12 = 2

1
-1
0

c11 = -6

1
1
0

0.1667
0.0000
-0.1667
0.0000

-2
6
-4
0

1
1
1

1
1
1

1
1
1

1
1
1

2
-1
0

4
1
0

1
2
-1
0

1
4
1
0

1
8
-1
0

c21=

c22=

c23=

c24=

6
-12
-6
12
-0.1667
0.5000
-0.3333
0.0000

c24 = -12

(-1)4+1

c23 = 6

(-1)3+1

c22 = 0

(-1)2+1

12

c21 = 6

(-1)1+1

1
1
1

1
1
1
1

Table 2.3. Example 2.6 Method of adjoints

1
8
-1

1
8
0

1
-1
0

8
-1
0

1
1
1

2
-1
0

10
26
2
4

1
1
1
1
1
1
1
1
1

0.5000
1.0000
-0.5000
1.0000

c34 = -6

1
2
-1

c33 = -4

1
2
0

c32 = -2

1
-1
0

c31 = 12

2
-1
0

8
-1
0

(-1)4+2

(-1)3+2

(-1)2+2

(-1)1+2

1
1
1
8
-1
0

1
8
-1

c44 = 12

c34=

c43 = 0

c33=

1
8
0

1
-1
0

c42 = 0

c32=

1
12

c41 = 0

c31=

1
2
3
4

1
4
1

1
4
0

1
1
0

4
1
0

8
-1
0

1
1
1

1
1
1

1
1
1

1
1
1

4
1
0

(-1)4+3

(-1)3+3

(-1)2+3

(-1)1+3

1
1
1

c44=

c43=

c42=

c41=

1
8
-1

1
8
0

1
-1
0

8
-1
0

1
4
1

1
4
0

1
1
0

4
1
0

8
-1
0

1
2
-1

1
2
0

1
-1
0

2
-1
0

4
1
0

(-1)4+4

(-1)3+4

(-1)2+3

(-1)1+4

2
-1
0

62NUMERICAL STRUCTURAL ANALYSIS

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 63

x1 + x2 + x3 + x4 = 10
8 x1 + 4 x2 + 2 x3 + x4 = 26
x1 + x2 x3 + x4 = 2
0 x1 + 0 x2 + 0 x3 + x4 = 4

[C ]T

[A ] =

and ci j = (1)i + j Aji

The determinants are shown by row expansion for the 4 4 matrix and by
the basket weave for all the 3 3 matrices in Table 2.3.
The last step in Table 2.3 is to multiply the invert of A, [A]1, by the
constant vector, [C], to get the final solution vector, [x].

2.6GAUSSIAN ELIMINATION METHOD


This method is named for Carl Friedrich Gauss who developed it in 1670
(Newton 1707). It was referenced by the Chinese as early as 179 (Anon
179). Gaussian elimination is a method for solving matrix equations by
composing an augmented matrix, [A|C], and then utilizing elementary
row operations to reduce this matrix into upper triangular form, [U|D].
Theelementary row operations used to reduce the matrix into upper triangular form consist of multiplying an equation by a scalar, or adding two
equations to form another equation. The equation used to eliminate terms
in other equations is referred to as the pivot equation. The coefficient of
the pivot equation that lies in the column of terms to be eliminated is
called the pivot coefficient or pivot element.
If this coefficient of the pivot element is zero, the pivot row must be
exchanged with another row. This exchange is called partial pivoting. If the
row with the largest element in the pivot column is exchanged to the pivot
row, accuracy is increased. Partial pivoting is when the rows are interchanged
and full pivoting is when both the rows and columns are reordered to place
a particular element in the diagonal position prior to a particular operation.
Whenever partial pivoting is utilized, the determinant changes sign with
each pivot unless done before reduction starts. However, the value of the
determinant of the matrix is not affected by elementary row operations.
Once the matrix is reduced to an equivalent upper triangular matrix,
the solutions are found by solving equations by back substitution. Thefollowing is the reduction procedure in algorithmic form:
aijk = aijk 1

akjk1
a

k 1
kk

(a )
k 1
ik

k + 1 j m
where

k +1 i n

64NUMERICAL STRUCTURAL ANALYSIS

With the following variables:


ak1 original elements
ak new elements
i row (n)
j column (m)
k pivotal row number
The following is the back substitution procedure in algorithmic form:

xn =

anm
ann
aim j = i +1aij x j
n

xi =

aii

where i = n 1, n 2, ,1

Example 2.7 Gaussian elimination method


Find the solution set to the following nonhomogeneous linear algebraic
equations using Gaussian elimination.
x1 + x2 + x3 + x4 = 10
8 x1 + 4 x2 + 2 x3 + x4 = 26
x1 + x2 x3 + x4 = 2
0 x1 + 0 x2 + 0 x3 + x4 = 4
The reduction process is shown in Table 2.4. The numbers to the right of
each row outside the augmented matrix, [A|C], are the reduction multipliers. The pivot row is multiplied by these numbers to reduce the rows below
the pivot row. As an example, the first row is multiplied by 8 and added
to the second row, producing a zero in the first column of the second row.
Then, the first row is multiplied by 1 and added to the third row, producing a
zero in the first column of the third row. Last, the first row is multiplied by0
and added to the last row, producing a zero in the first column of the last
row. The result is the reduction of all the values below the pivot element in
the first column to zero. The second column is then reduced to zeros below
the pivot element, and lastly the third column is reduced to zeros below the
pivot element. The solution vector, [x], is also shown.
These values for x are solved after the final reduction. From row four,
the following equation can be written:
1x4 = 4 x4 = 4

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 65

Table 2.4. Example 2.7 Gaussian elimination


A|C

A|C

A|C

A|C

1
8
1
0

1
4
1
0

1
2
1
0

1
1
1
1

10
26
2
4

1
0
0
0

1
4
2
0

1
6
0
0

1
7
2
1

10
54
12
4

8
1
0

1
0
0
0

1
4
0
0

1
6
3
0

1
7
1.5
1

10
54
15
4

0.5
0

1
0
0
0

1
4
0
0

1
6
3
0

1
7
1.5
1

10
54
15
4

1
2
3
4

From row three, the following equation is valid:


3 x3 1.5 x4 = 15
Substituting x4 = 4 into this equation yields:
3 x3 1.5 ( 4) = 15 3 x3 = 9 x3 = 3
Row two produces the following equation:
4 x2 6 x3 7 x4 = 54
Substituting x3 = 3 and x4 = 4 into this equation yields:
4 x2 6 (3) 7 ( 4) = 54 4 x2 = 8 x2 = 2

66NUMERICAL STRUCTURAL ANALYSIS

And, from row one the following equation:


1x1 + 1x2 + 1x3 + 1x4 = 10
Substituting x2 = 2, x3 = 3, and x4 = 4 into this equation yields:
1x1 + 1( 2) + 1(3) + 1( 4) = 10 x1 = 1
It should be noted that the product the diagonal values of any triangular
matrix is the determinant. In this case, the determinant is |A| = (1)(4)(3)
(1) = 12.
Example 2.8 Gaussian elimination method
Determine the solution to the following set of equations using Gaussian
elimination and back substitution using partial pivoting. Include a determinant check for uniqueness.
x1 + x2 + x3 + x4 = 10
8 x1 + 4 x2 + 2 x3 + x4 = 26
x1 + x2 x3 + x4 = 2
3 x1 + 2 x2 + 1x3 + 0 x4 = 10
The reduction process is shown in Table 2.5. Again, the numbers to the
right of each row outside the augmented matrix, [A|C], are the reduction
multipliers. Also noted is the partial pivoting. Note that the second row
has the largest number in the first column. Therefore, that row is swapped
with the first row, placing the largest element in the pivot element position. Reduction is then performed on the first column. After reduction of
the first column, the largest number in the second column is in the third
row. That row is swapped with the second row, placing the largest number
in the pivot position. Reduction is then performed on the second column.
The third column does not require partial pivoting, since the largest number in the third column is already in the pivot position.
Since two partial pivots were performed, the product of the diagonal
must be multiplied by (1)2 to achieve the correct sign on the determinant,
|A| = (8)(1.5)(1)(1) (1)2 = 12. Back substitution is performed to determine the solution vector, [x], which is shown in Table 2.5.

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 67

Table 2.5. Example 2.8 Gaussian elimination

10

26

10

26

Swap

10

Pivot

10

26

0.5

0.75

0.88

6.75

1.13

5.25

0.125

0.25

0.375

1.5

0.5

0.8
0.25
2

0.4

0.125

1.5

0.5

0.75

0.5

0.25

26

1.5

0.8

1.13

5.25

0.5

0.333

0.5

0.8

1.5

0.333

26

1.5

0.8

1.13

5.25

0.5

0.500

12

* (1)2

0.8

26

1.13

5.25

Swap

0.88

6.75

Pivot

0.4

0.25

1
x

2
3
4

68NUMERICAL STRUCTURAL ANALYSIS

2.7GAUSSJORDAN ELIMINATION METHOD


The GaussJordan elimination method is a variation of the Gaussian
elimination method in which an unknown is eliminated from all equations except the pivot during the elimination process. This method was
described by Wilhelm Jordan in 1887 (Clasen 1888). When an unknown is
eliminated, it is eliminated from equations preceding the pivot equation as
well as those following the pivot equation. The result is a diagonal matrix
and eliminates the need for back substitution. In fact, if the pivot elements
are changed to ones by dividing each row by the pivot element, the last
column will contain the solution.
The disadvantage to the GaussJordan elimination method is that two
matrices are required for elimination, however, they do get smaller as the
process continues. The previous pivot is moved to the bottom, with a new
pivot on top. The GaussJordan process is shown in Example 2.9.
Example 2.9 GaussJordan elimination method
Determine the solution to the following set of equations using Gauss
Jordan elimination.
2 x1 2 x2 + 5 x3 = 13

(2.1a)

2 x1 + 3 x2 + 4 x3 = 20

(2.1b)

3 x1 x2 + 3 x3 = 10

(2.1c)

The first step is to divide the first equation of the set by the coefficient of
the first unknown in that equation, 2. Equation 2.2a is then multiplied by
the corresponding coefficient of that unknown of Equations 2.1b and 2.1c
to give the following:
x1 x2 +

5
13
x3 =
2
2

2 x1 2 x2 + 5 x3 = 13
3 x1 3 x2 +

15
39
x3 =
2
2

(2.2a)
(2.2b)
(2.2c)

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 69

Next, Equation 2.2b is subtracted from 2.1b and becomes 2.3a. E


quation
2.2c is subtracted from 2.1c and becomes 2.3b. Equation 2.2a now
becomes 2.3c, thereby moving to the bottom.
5 x2 x3 = 7
2 x2

(2.3a)

19
9
x3 =
2
2

(2.3b)

5
13
x3 =
2
2

(2.3c)

x1 x2 +

Now, Equation 2.3a is divided by the first unknown in that equation, 5,


and the new Equation 2.4a is multiplied by the corresponding coefficient
of that unknown from the two other equations (2.3b and 2.3c) to yield
Equations 2.4b and 2.4c.
1
7
x2 x3 =
5
5

(2.4a)

14
2
x3 =
5
5

(2.4b)

1
7
x2 + x3 =
5
5

(2.4c)

2 x2

Just as the previous cycle, Equation 2.4b is subtracted from 2.3b and
becomes 2.5a. Equation 2.4c is subtracted from 2.3c and becomes 2.5b.
Equation 2.4a now becomes 2.5c.

123
41
x3 =
10
10

(2.5a)

23
79
x3 =
10
10

(2.5b)

x1 +

1
7
x2 x3 =
5
5

(2.5c)

70NUMERICAL STRUCTURAL ANALYSIS

Equation 2.5a is divided by the first unknown in that equation, 41/10


and the new Equation 2.6a is multiplied by the corresponding coefficient
of that unknown from the two other equations (2.5b and 2.5c) to yield
Equations 2.6b and 2.6c.
x3 = 3

(2.6a)

23
69
x3 =
10
10

(2.6b)

3
1
x3 =
5
5

(2.6c)

Equation 2.6b is subtracted from 2.5b and becomes 2.7a. Equation 2.6c is
subtracted from 2.5c and becomes 2.7b. Equation 2.6a now becomes 2.7c.
All the unknowns are found with the following solution:
x1 = 1

(2.7a)

x2 = 2

(2.7b)

x3 = 3

(2.7c)

The same result may be obtained by working with just the coefficients
and constants of the equations. Given the same equations, the following
augmented matrix is valid:
2 x1 2 x2 + 5 x3 = 13
2 x1 + 3 x2 + 4 x3 = 20
3 x1 x2 + 3 x3 = 10
2 2 5 13
[ A] = 2 3 4 20
3 11 3 10
An augmented matrix, [B], is established from the following algorithm:

bi 1, j 1 = aij

a1 j ai1
a11

1< i n

where 1 < j m
a 0
11

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 71

Also, the final row of the new matrix is found by:

bn , j 1 =

a1 j
a11

1 < j m
where

a11 0

where,
a
b
i
j

elements in old matrix A


elements in new matrix B
row number in old matrix A (n)
column number of old matrix A (m)

For example,
b2, 2 = a2, 2

a1, 2 a2,1
a11

= 3

2 ( 2)
=5
2

From these equations, column 1 is reduced.

5 1

9
[ B ] = 2 2

5
1

19

2
13

Following the process again to reduce column 2:


123
41
10 10

23
79

[C ] = 10 10

7
1
5
5
Finally, one more cycle reduces column 3 and obtains the solution:
1
D
=
[ ] 2
3
It is known that applying these equations to a number of simultaneous
equations, n, will produce the same number of matrices, n, to achieve the
solution.

72NUMERICAL STRUCTURAL ANALYSIS

2.8IMPROVED GAUSSJORDAN ELIMINATION


METHOD
In comparison to the GaussJordan elimination method, the improved
GaussJordan elimination method uses the same space for both the
A and B arrays. This is beneficial if the amount of space available on
the computer is limited. The algorithm for this improved method is as
follows:
akj

akj =

akk

aij = aij aik akj k = 1, 2, 3 n except i k


1 i n

k + 1 j n + 1
where,
a
a
i
j
k

original elements
new elements
row (n)
column (m)
pivotal row number

In other words, normalize the matrix then utilize partial pivoting to reduce
the matrix. However, there is no need to reduce the elements under the
pivot. Reducing up and down is easier without a need to reorder the rows.
This is how Example 2.10 is performed.

Example 2.10 Improved GaussJordan elimination method


Determine the solution to the following set of equations using improved
GaussianJordan elimination. Include a determinant check for
uniqueness.
x1 + x2 + x3 + x4 + x5 = 8
16 x1 + 8 x2 + 4 x3 + 2 x4 + 1x5 = 44
x1 x2 + x3 x4 + x5 = 2
81x1 27 x2 + 9 x3 3 x4 + x5 = 44
16 x1 8 x2 + 4 x3 2 x4 + x5 = 8

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 73

1
1

16 8
A C = 1 1

81 27
16 8

1 1
4 2
1 1
9 3
4 2

1 8

1 44
1 2

1 44
1 8

From partial pivoting, the first row can be swapped with the fourth row to
form the following matrix:
81 27

16 8
A C = 1 1

1
1
16 8

9 3 1 44

4 2 1 44
1 1 1 2

1 1 1 8
4 2 1 8

Now elimination may be performed as shown in Table 2.6. Note that for
each column reduction, elements are reduced to zero below and above
the pivot position. Once it reduces to a diagonal matrix, the solution is
foundby dividing each row by the pivot element. The determinant is
found as |A| = (81)(13.3333)(1)(1.3333) (2) = 2880.

2.9CHOLESKY DECOMPOSITION METHOD


Cholesky decomposition is also known as Crouts method or matrix
factorization. This method was discovered by Andr-Louis Cholesky
(Commandant Benoit 1924). Cholesky decomposition changes the original augmented equation to an equivalent upper and lower triangular set.
If a set of three simultaneous equations exist, they can be represented as
follows:
a11
a
21
a31

a12
a22
a32

a13 x1 c1
a23 x2 = c2
a33 x3 c3

If [A] represents the coefficient matrix, [x] represents the column matrix of
the unknowns, and [C] represents the column matrix of the constants, the
previous can be expressed as the following equation:

74NUMERICAL STRUCTURAL ANALYSIS

Table 2.6. Example 2.10 Improved GaussianJordan elimination


method

81
16
1
1
16

-27
8
-1
1
-8

9
4
1
1
4

-3
2
-1
1
-2

1
1
1
1
1

44
44
2
8
8

81
0
0
0
0

-27
13.3333
-0.6667
1.3333
-2.6667

9
2.2222
0.8889
0.8889
2.2222

-3
2.5926
-0.9630
1.0370
-1.4074

1
0.8025
0.9877
0.9877
0.8025

44
35.3086
1.4568
7.4568
-0.6914

81
0
0
0
0

0
13.5
13.3333 2.2222
0
1
0
0.6667
0
2.6667

2.25
2.5926
-0.8333
0.7778
-0.8889

2.625
0.8025
1.0278
0.9074
0.9630

115.5
35.3086
3.2222
3.9259
6.3704

0
13.3333
0
0
0

0
0
1
0
0

13.5
4.4444
-0.8333
1.3333
1.3333

-11.25
-1.4815
1.0278
0.2222
-1.7778

72
28.1481
3.2222
1.7778
-2.2222

-13.5000
-2.2222

81
0
0
0
0
81
0
0
0
0

0
13.3333
0
0
0

0
0
1
0
0

0
0
0
1.33333
0

-13.5
-2.2222
1.1667
0.2222
-2

54
22.2222
4.3333
1.7778
-4

-10.1250
-3.3333
0.6250

81
0
0
0
0

0
13.3333
0
0
0

0
0
1
0
0

0
0
0
1.33333
0

0
0
0
0
-2

1
2
2
1
2

-2880.00

-0.1975
-0.0123
-0.0123
-0.1975
2.0250
0.0500
-0.1000
0.2000

-0.6667
-2.6667

-1.0000

81
-6.7500
26.6667 -1.1111
2
0.5833
1.33333 0.1111
-4

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 75

[ A][ x ] = [C ]

or

[ A][ x ] [C ] = 0

If the original system of equations is reduced into an equivalent system in


upper triangular form, the following is true:
1 u12
0 1

0 0

u13 x1 d1
u23 x2 = d 2
1 x3 d3

[U ][ x ] [ D ] = 0
Also, a lower triangular matrix exists, such that, when the first set is
pre-multiplied by [L], the result is the second set as follows:

l11 0
l
21 l22
l31 l32

0
0
l33

[ L ]([U ][ x ] [ D ]) = [ A][ x ] [C ]
[ L ][U ] = [ A]

and

[ L ][ D ] = [C ]

In matrix form, it looks as follows:


l11 0
l
21 l22
l31 l32

0 1 u12
0 0 1
l33 0 0

u13
u23
1

d1 a11
d 2 = a21
d3 a31

a12
a22
a32

a13
a23
a33

c1
c2
c3

The order of the solution process is as follows with each producing an


equation involving only one unknown:
1. Obtain column 1 of [L] by multiplying each row of [L] by column1
of [U] to get column 1 of [A]. That is, use a11, a21, a31 to get l11,
l21, l31.
2. Obtain row 1 of [U] by multiplying row 1 of [L] times each column
of [U] to get row 1 of [A], excluding column 1 of [U]. That is, use
a12, a13, c1 to get u12, u13, d1.

76NUMERICAL STRUCTURAL ANALYSIS

3. Obtain column 2 of [L] by multiplying each row of [L] times column


2 of [U] to get column 2 of [A], excluding row 1 of [L], That is, use
a22, a32 to get l22, l32.
4. Obtain row 2 of [U] by multiplying row 2 of [L] times each column
of [U] to get row 2 of [A], excluding columns 1 and 2 of [U]. That
is, use a23, c2 to get u23, d2.
5. Obtain column 3 of [L] by multiplying each row of [L] times column 3 of [U] to get column 3 of [A], excluding row 1 and 2 of [L].
That is, use a33 to get l33.
6. Obtain row 3 of [U] by multiplying row 3 of [L] times each column
of [U] to get row 3 of [A], excluding columns 1, 2, and 3 of [U].
That is, use c3 to get d3.
All of these arithmetic operations can be done with an algorithm as
follows:
i = 1, 2, 3, , n
li1 = ai1 For

j =1

a1 j
j = 2, 3, 4, , n + 1
u1 j =
For

i =1
a11

j = 2, 3, 4, , n

For i = j , j + 1, j + 2, , n
( for each value of j )

j 1

li , j = aij lik ukj


k =1

aij k =1lik ukj


i 1

ui , j =

lii

i = 2, 3, 4, , n

For j = i, i + 1, i + 2, , n + 1
( for each value of i )

xn = un , n +1
xi = ui , n +1

j = i +1

ij

xj

For {i = n 1, n 2, n 3, , 1}

An improved Cholesky decomposition scheme can also be used with only


one matrix in the process as follows:
a1 j =

a1 j
a11

For

{ j = 2, 3, 4, , n + 1}

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 77

j 1

ai , j = aij aik akj


k =1

j = 2, 3, 4, , n

For i = j , j + 1, j + 2, , n
( for each value of j )

i = 2, 3, 4, , n

For j = i, i + 1, i + 2, , n + 1
( for each value of i )

aij k =1aik akj


i 1

ai , j =

aii

Example 2.11 Cholesky decomposition method


Find the solution set to the following non-homogeneous linear algebraic
equations using Cholesky decomposition.
x1 + x2 + x3 + x4 + x5 = 8
16 x1 + 8 x2 + 4 x3 + 2 x4 + 1x5 = 44
x1 x2 + x3 x4 + x5 = 2
81x1 27 x2 + 9 x3 3 x4 + x5 = 44
16 x1 8 x2 + 4 x3 2 x4 + x5 = 8

l11 0
l
21 l22
l31 l32

0 1 u12
0 0 1
l33 0 0

u13
u23
1

d1 a11
d 2 = a21
d3 a31

a12
a22
a32

a13
a23
a33

c1
c2
c3

The matrix in augmented form is shown in Table 2.7.

Table 2.7. Example 2.11 Cholesky decomposition


method
1
A

16

-1

44

-1

81

-27

-3

44

16

-8

-2

78NUMERICAL STRUCTURAL ANALYSIS

The reduced lower triangular matrix is shown in Table 2.8.


Table 2.8. Example 2.11 Cholesky decomposition method
1
L

16

-8

-2

81

-108

90

60

16

-24

24

12

-2

The upper triangular matrix is found at the same time and is shown in
Table 2.9.
Table 2.9. Example 2.11 Cholesky decomposition method

U|D

1.5

1.75

1.875

10.5

0.5

1.25

0.167

1.333

Finally, the solution is calculated from the [U|D] matrix using back substitution and shown in Table 2.10.
Table 2.10. Example 2.11 Cholesky decomposition method
1
2
x

2
1
2

2.10 ERROR EQUATIONS


Error equations are intended to increase the accuracy in which the roots
of simultaneous equations are determined by reducing the error due to
rounding off. Consider the set of equations as follows:

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 79

a11 x1 + a12 x2 + a13 x3 + + a1n xn = C1


a21 x1 + a22 x2 + a23 x3 + + a2 n xn = C2
a31 x1 + a32 x2 + a33 x3 + + a3n xn = C3

an1 x1 + an 2 x2 + an 3 x3 + + ann xn = Cn
If the approximate roots x1, x2, x3, xn have been obtained by elimination, upon substitution into the equations the constants C1, C2, C3, Cn
are found as follows:
a11 x1 + a12 x2 + a13 x3 + + a1n xn = C1
a21 x1 + a22 x2 + a23 x3 + + a2 n xn = C2
a31 x1 + a32 x2 + a33 x3 + + a3n xn = C3

an1 x1 + an 2 x2 + an 3 x3 + + ann xn = Cn
If x1, x2, x3 , xn are the corrections that must be added to the
approximate root to obtain the exact root values x1, x2, x3,, xn, the
following is utilized:
x1 = x1 + x1
x2 = x2 + x2
x3 = x3 + x3

xn = xn + xn
If we substitute these expressions for the exact root, we obtain the following:
a11 ( x1 + x1 ) + a12 ( x2 + x2 ) + a13 ( x3 + x3 ) + + a1n ( xn + xn ) = C1
a21 ( x1 + x1 ) + a22 ( x2 + x2 ) + a23 ( x3 + x3 ) + + a2 n ( xn + xn ) = C2
a31 ( x1 + x1 ) + a32 ( x2 + x2 ) + a33 ( x3 + x3 ) + + a3n ( xn + xn ) = C3

an1 ( x1 + x1 ) + an 2 ( x2 + x2 ) + an 3 ( x3 + x3 ) + + ann ( xn + xn ) = Cn

80NUMERICAL STRUCTURAL ANALYSIS

If these equations are subtracted from the approximate equations, the


following is obtained:
a11 x1 + a12 x2 + a13 x3 + + a1n xn = C1 C1 = e1
a21 x1 + a22 x2 + a23 x3 + + a2 n xn = C2 C2 = e2

a31 x1 + a32 x2 + a33 x3 + + a3n xn = C3 C3 = e3




an1 x1 + an 2 x2 + an 3 x3 + + ann xn = Cn Cn = e
This shows that the corrections, xs, can be obtained by replacing the
constant vector of the solution with the difference of the constant vectors,
(CCs)s, and applying reduction to find the error. These are then added
to the approximate solution and the process is repeated until accuracy is
achieved.
Example 2.12 Error equations
Determine the solution to the following set of equations using any
GaussJordan elimination, but only carry two decimals of accuracy
(i.e., x.xx) then apply error equations to increase accuracy.
2.11x1 + 2.11x2 3.04 x3 + 1.11x4 = 1.65
0.02 x1 + 1.23 x2 + 2.22 x3 + 1.02 x4 = 13.18
0.14 x1 0.06 x2 + 1.21x3 1.08 x4 = 0.67
1.32 x1 + 0.20 x2 + 0.000 x3 + 3.90 x4 = 17.32
The process of three complete cycles is shown in Tables 2.112.13.

2.11 MATRIX INVERSION METHOD


The solution to a set of linear equations can be achieved by using any reduction technique on the coefficient matrix augmented with the identity matrix.
a11
a
21
[ A | I ] = a31


an1

a12
a22
a32

an 2

a13  a1n
a23  a2 n
a33  a3n
  
an 3  ann

1 0 0 
0 1 0 
0 0 1 
   
0 0 0 

0
0
0


1

From here, the coefficient matrix is reduced until the identity matrix is
on the left and the original identity on the right becomes the invert of A.

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 81

Table 2.11. Example 2.12 Error equations


2.11
A|C

A|C

A|C

A|C

A|C

-0.02

-3.04

2.22
1.21
0.00

1.11
1.02

1.65
13.18

-1.08

-0.67

3.90

17.32
1.65
13.20
-0.78
16.29

0.14
1.32

-0.06

2.11
0.00
0.00
0.00

2.11
1.25
-0.20
-1.12

-3.04

2.19
1.41
1.90

1.11
1.03
-1.15
3.21

2.11
0.00

0.00
1.25

-6.74

-0.63

-20.63

2.19

1.03

13.20

0.00
0.00

0.00
0.00

1.76
3.86

-0.99

4.13

1.33
28.12

2.11
0.00
0.00
0.00

0.00
1.25
0.00
0.00

0.00
0.00
1.76
0.00

-4.42

-15.54

2.26
6.30

11.55
1.33
25.20

2.11
0.00
0.00
0.00

0.00
1.25
0.00
0.00

0.00
0.00
1.76
0.00

0.00
0.00
0.00
6.30

2.14
2.51
5.29
25.20

1.01
2.01
3.01
4.00
-0.01

e1

2.11
1.23

-0.03
-0.01
-0.02

0.20

C1

-0.99

1.66
13.21
-0.66
17.34

82NUMERICAL STRUCTURAL ANALYSIS

Table 2.12. Example 2.12 Error equations


A|C

A|C

A|C

A|C

A|C

2.11

2.11

-3.04

1.11

-0.01

-0.02

1.23

2.22

1.02

-0.03

0.14

-0.06

1.21

-1.08

-0.01

1.32

0.20

0.00

3.90

-0.02

2.11

2.11

-3.04

1.11

-0.01

0.00

1.25

2.19

1.03

-0.03

0.00

-0.20

1.41

-1.15

-0.01

0.00

-1.12

1.90

3.21

-0.01

2.11

0.00

-6.74

-0.63

0.04

0.00

1.25

2.19

1.03

-0.03

0.00

0.00

1.76

-0.99

-0.01

0.00

0.00

3.86

4.13

-0.04

2.11

0.00

0.00

-4.42

0.00

0.00

1.25

0.00

2.26

-0.02

0.00

0.00

1.76

-0.99

-0.01

0.00

0.00

0.00

6.30

-0.02

2.11

0.00

0.00

0.00

-0.01

0.00

1.25

0.00

0.00

-0.01

0.00

0.00

1.76

0.00

-0.01

0.00

0.00

0.00

6.30

-0.02

0.00
x

-0.01

1.01
Dx1

-0.01

3.00

0.00

4.00

1.67
C2

2.00

13.18

-0.02
e2

0.00

-0.67

0.00

17.33

-0.01

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 83

Table 2.13. Example 2.12 Error equations


A|C

A|C

A|C

A|C

A|C

2.11

2.11

-3.04

1.11

-0.02

-0.02

1.23

2.22

1.02

0.00

0.14

-0.06

1.21

-1.08

0.00

1.32

0.20

0.00

3.90

-0.01

2.11

2.11

-3.04

1.11

-0.02

0.00

1.25

2.19

1.03

0.00

0.00

-0.20

1.41

-1.15

0.00

0.00

-1.12

1.90

3.21

0.00

2.11

0.00

-6.74

-0.63

-0.02

0.00

1.25

2.19

1.03

0.00

0.00

0.00

1.76

-0.99

0.00

0.00

0.00

3.86

4.13

0.00

2.11

0.00

0.00

-4.42

-0.02

0.00

1.25

0.00

2.26

0.00

0.00

0.00

1.76

-0.99

0.00

0.00

0.00

0.00

6.30

0.00

2.11

0.00

0.00

0.00

-0.02

0.00

1.25

0.00

0.00

0.00

0.00

0.00

1.76

0.00

0.00

0.00

0.00

0.00

6.30

0.00

0.01
x

0.00

1.00
Dx2

0.00

3.00

0.00

4.00
0.00

1.67
C3

2.00

13.18

e3

0.00

-0.67

0.00

17.32

0.00

84NUMERICAL STRUCTURAL ANALYSIS

0
1
I | A = 0


0

0
1
0

0

0
0
1

0

0 a111
0 a211
1
0 a31
 
1 an11







a121
1
a22
1
a32

an21

a131 
1
a23

1
a33 
 
an31 

a1n1

a2n1
a3n1


1
ann

If partial pivoting is used during the reduction, the columns in the


invert must be swapped back in reverse order as the rows were swapped
during reduction. The constant vector must also be reordered in the
same way.
Example 2.13 Matrix inversion
Determine the solution to the following set of equations using inversionin-place (improved GaussJordan). Include partial pivoting during the
reduction.
2.11x1 + 2.11x2 3.04 x3 + 1.11x4 = 1.65
0.02 x1 + 1.23 x2 + 2.22 x3 + 1.02 x4 = 13.18
0.14 x1 0.06 x2 + 1.21x3 1.08 x4 = 0.67
1.32 x1 + 0.20 x2 + 0.000 x3 + 3.90 x4 = 17.32

The matrix is shown in Table 2.14 in augmented form and GaussJordan


elimination is performed for the first two columns.
Table 2.14. Example 2.13 Matrix inversion method
2.11

2.11
1.23

A|C = -0.02
0.14
1.32

-0.06
0.20

2.11

2.11

A|I = -0.02
0.14
1.32

-0.06
0.20

1.23

-3.04
2.22
1.21
0.00
-3.04
2.22
1.21
0.00

1.11
1.02

1.65
13.18

-1.08
3.90

-0.67
17.32

1.11

1.00

0.00

0.00

0.00

1.02

0.00
0.00
0.00

1.00
0.00
0.00

0.00
1.00
0.00

0.00
0.00
1.00

-1.08
3.90

(Continued)

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 85

Table 2.14.(Continued)
A|I

A|I

2.110
0.000
0.000

2.110
1.250
-0.200

-3.040
2.191
1.412

0.000

-1.120

1.902

2.1100
0.0000
0.0000

0.0000
1.2500
0.0000

0.0000

0.0000

1.110
1.031

1.000
0.009
-0.066

0.000
1.000
0.000

0.000
0.000
1.000

0.000
0.000
0.000

-1.154
3.206

-0.626

0.000

0.000

1.000

-6.7387 -0.6295 0.9840 -1.6880 0.0000


2.1912 1.0305 0.0095 1.0000 0.0000
1.7623 -0.9888 -0.0648 0.1600 1.0000
3.8651 4.1289 -0.6171 0.8960 0.0000

0.0000
0.0000
0.0000
1.0000

Note that partial pivoting should be performed and row three is now
swapped with row four. Columns 3 and 4 are then eliminated as shown in
Table 2.15.
Table 2.15. Example 2.13 Matrix inversion method
A|I

A|I

A|I

2.1100 0.0000 -6.7387 -0.6295

0.9840 -1.6880

0.0000

0.0000

0.0000 1.2500

2.1912

1.0305

0.0095

1.0000

0.0000

0.0000

0.0000 0.0000

3.8651

4.1289 -0.6171

0.8960

0.0000

1.0000

0.0000 0.0000

1.7623 -0.9888 -0.0648

0.1600

1.0000

0.0000

2.1100 0.0000

0.0000

6.5692 -0.0919 -0.1258

0.0000

1.7435

0.0000 1.2500

0.0000 -1.3102

0.0000 0.0000

3.8651

0.0000 0.0000

0.0000 -2.8714

0.2165 -0.2485

1.0000 -0.4560

2.1100 0.0000

0.0000

0.0000

0.4035 -0.6944

2.2878

0.0000 1.2500

0.0000

0.0000

0.2605

0.0000 0.0000

3.8651

0.0000 -0.3057

0.0000 0.0000

0.0000 -2.8714

0.3593

0.4920

0.0000 -0.5669

4.1289 -0.6171

0.8960

0.0000

1.0000

0.7003

0.6055 -0.4563 -0.3589


0.5386

0.2165 -0.2485

1.4380

0.3444

1.0000 -0.4560

Now the third and fourth columns are swapped back for the inverse
matrix shown in Table 2.16.
Table 2.16. Example 2.13 Matrix inversion method
I|A-1

1.0843

0.3319

1.0000

0.0000

0.0000

0.0000

0.1912 -0.3291

0.0000

1.0000

0.0000

0.0000

0.2084

0.0000

0.0000

1.0000

0.0000

-0.0791

0.1394

0.3720

0.0891

0.0000

0.0000

0.0000

1.0000

-0.0754

0.0866 -0.3483

0.1588

0.4844 -0.3651 -0.2871

(Continued)

86NUMERICAL STRUCTURAL ANALYSIS

Table 2.16.(Continued)
0.1912 -0.3291
A

-1

0.2084

0.3319

1.0843

0.4844 -0.2871 -0.3651

-0.0791

0.1394

0.0891

-0.0754

0.0866

0.1588 -0.3483

0.3720

Now the third and fourth rows are swapped in coefficient matrix and
the solution is found as [x]=[A]-1[C] in Table 2.17:
Table 2.17. Example 2.13 Matrix inversion method
1.65
C

13.18

1.65
C

13.18

1
x

-0.67

17.32

17.32

-0.67

2.12GAUSSSEIDEL ITERATION METHOD


Some methods such as the Gaussian elimination are not appropriate
when a sparse matrix exists. A matrix normally can be considered sparse
if approximately two-thirds or more of the entries in a matrix are zero.
The GaussSeidel iteration method was developed for such systems. The
method is named for Carl Friedrich Gauss and Philipp Ludwig von Seidel
(Gauss 1903). This method is an iteration method in which the last calculated values are used to determine a more accurate solution. Typically,
all unknown x values are assumed to be zero to begin the iteration. This
method mainly works best with a diagonal system in which the largest
values lie on the diagonal. The elastic stiffness matrix used to analyze
structures is a typical example of a diagonal system and will be presented
in Chapter 4. A diagonal system is sufficient, but not necessary to provide
convergence. During the process, each row is used to find a better approximation of the variable corresponding to the row using all other variables
as known.
Example 2.14 GaussSeidel lteration method
Determine the solution to the following set of equations using the
GaussSeidel iteration method with e = 0.01 and assume x1 = x2 = x3
= x4 = 0.

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 87

10 x1 + x2 + 2 x3 + x4 = 50
2 x1 + 10 x2 + x3 + 2 x4 = 63
x1 + 2 x2 + 10 x3 + x4 = 67
2 x1 + x2 + x3 + 10 x4 = 75
Note that this is a diagonal system with 10s on the diagonal and all other
coefficients are much less. Begin the iteration by setting x1 = x2 = x3 = x4 = 0
and solving for each of the unknowns using the corresponding equation in
a top down order.
10 x1 + 0 + 2 ( 0) + 0 = 50 x1 = 5.000

2 (5.000) + 10 x2 + 0 + 2 ( 0) = 63 x2 = 5.300

5.000 + 2 (5.300) + 10 x3 + 0 = 67 x3 = 5.140

2 (5.000) + 5.300 + 5.140 + 10 x4 = 75 x4 = 5.456

After completing the first cycle, start with the first equation using the new
values and find a closer approximation for each unknown. Also, check the
difference between the new values and the previous values to determine if
the desired accuracy is achieved.
10 x1 + 5.300 + 2 ( 5.140 ) + 5.456 = 50 x1 = 2.896 and x1 = 2.104

2 ( 2.896 ) + 10 x2 + 5.140 + 2 ( 5.456 ) = 63 x2 = 4.116 and x2 = 1.184

2.896 + 2 ( 4.116 ) + 10 x3 + 5.456 = 67 x3 = 5.042 and x3 = 0.098


2 ( 2.896 ) + 4.116 + 5.042 + 10 x4 = 75 x4 = 6.005 and x4 = 0.549

None of the values of Dx are less than e = 0.01, so the process is repeated.
Table 2.18 shows the entire process to convergence. The process can be
stopped when each value has changed less than e or when a cycle results
in each value changing less than e.

2.13 EIGENVALUES BY CRAMERS RULE


A homogeneous equation is one where all the constants on the right-hand
side of the equal sign are zero. The typical set of n homogeneous equations
with n unknown solution sets is as follows:

88NUMERICAL STRUCTURAL ANALYSIS

a11 x1 + a12 x2 + a13 x3 + + a1n xn = 0


a21 x1 + a22 x2 + a23 x3 + + a2 n xn = 0
a31 x1 + a32 x2 + a33 x3 + + a3n xn = 0


an1 x1 + an 2 x2 + an 3 x3 + + ann xn = 0
This can be written in matrix form, [A][x]=[0], as follows:

Table 2.18. Example 2.14 GaussSeidel iteration method


x1=
x2=
x3=
x4=

5.000
5.300
5.140
5.456

x1=
x2=
x3=
x4=

5.000
5.300
5.140
5.456

x1=
x2=
x3=
x4=

2.896
4.116
5.042
6.005

x1=
x2=
x3=
x4=

-2.104
-1.184
-0.098
0.549

x1=
x2=
x3=
x4=

2.980
3.999
5.002
6.004

x1=
x2=
x3=
x4=

0.083
-0.117
-0.040
-0.001

<epsilon

x1=
x2=
x3=
x4=

2.999
3.999
5.000
6.000

x1=
x2=
x3=
x4=

0.020
0.000
-0.002
-0.004

<epsilon
<epsilon
<epsilon

x1=
x2=
x3=
x4=

3.000
4.000
5.000
6.000

x1=
x2=
x3=
x4=

0.001
0.001
0.000
0.000

<epsilon
<epsilon
<epsilon
<epsilon

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 89

a11
a
21
a31


an1

a12
a22
a32

an 2

a13
a23
a33

an 3







a1n x1 0
a2 n x2 0

a3n x3 = 0

  0
ann xn 0

Let us consider the solution of eigenvalue problems. For any square


matrix, [A], the determinant equation |AlI|=0 is a polynomial equation of
degree n unknowns in the variable l. In other words, there are exactly n
roots that satisfy this equation. These roots are known as eigenvalues of A.

( a11 l ) x1 + a12 x2 + a13 x3 + + a1n xn = 0


a21 x1 + ( a22 l ) x2 + a23 x3 + + a2 n xn = 0
a31 x1 + a32 x2 + ( a33 l ) x3 + + a3n xn = 0

an1 x1 + an 2 x2 + an 3 x3 + + ( ann l ) xn = 0
Converting these equations into matrix form, [AlI][x]=0:
( a11 l )

a21
a31

a
n1

a12
(a22 l )
a32

an 2

a13
a23
(a33 l )

an 3







a1n x1 0

a2 n x2 0
a3n x3 = 0


 0
(ann l ) xn 0

The non-trivial solution exists if the determinant of the coefficient


matrix is zero. We use this so that Cramers rule can be used to find the
eigenvalues. Example 2.15 shows the process of determining the eigenvalues by Cramers rule.
Example 2.15 Eigenvalues by Cramers rule
Determine the eigenvalues for the following set of equations using
Cramers rule.
0 x1 + 2 x2 + 3 x3 = 0
10 x1 1x2 + 2 x3 = 0
2 x1 + 4 x2 + 7 x3 = 0

90NUMERICAL STRUCTURAL ANALYSIS

(0 l )
10
2

2
( 1 l )
4

3
2
=0
(7 l )

Solve the determinant by the basket weave method.

(0 l )

2
( 1 l )
4

10
2

3
2
(7 l )

(0 l )
10
2

2
( 1 l )
4

( 0 l ) ( 1 l ) ( 7 l ) + ( 2)( 2) ( 2) + (3) ( 10)( 4)

(3) ( 1 l ) ( 2) ( 0 l )( 2)( 4) ( 2) ( 10) ( 7 l )

l 6l 2 + 11l 6 = 0
3

The solution to the cubic equation can be found by many of the methods
from Chapter 1 and represent the eigenvalues l = 1, 2, and 3.

2.14FADDEEVLEVERRIER METHOD
The FaddeevLeverrier method is a polynomial method used to find the
eigenvalues. The method is named for Dmitrii Konstantinovich Faddeev
and published by Urbain Jean Joseph Le Verrier in 1840 (Le Verrier
1839). From linear algebra, the trace of a matrix is the sum of the diagonal terms. The process for determining the characteristic polynomial is
as follows:

( 1)n (ln p1ln 1 p2 ln 2 p3 ln 3 pn ) = 0


p1 = tr [ B1 ]
[ B1 ] = [ A]
1
tr [ B2 ]
2
1
p3 = tr [ B3 ]
3

[ B ] = [ A]([ B ] p [ I ])
2

p2 =

[ B ] = [ A]([ B ] p [ I ])
3


pk =

1
tr [ Bk ]
k

pn =

1
tr [ Bn ]
n

[ B ] = [ A]([ B ] p [ I ])
k

k 1

k 1

[ B ] = [ A]([ B ] p [ I ])
n

n 1

n 1

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 91

Example 2.16 shows the process of determining the characteristic polynomial and the eigenvalues by the FaddeevLeverrier method.
Example 2.16 FaddeevLeverrier method
Determine the eigenvalues for the following set of equations using the
FaddeevLeverrier method.
x1 + 2 x2 + 3 x3 = 0
10 x1 + 0 x2 + 2 x3 = 0
2 x1 + 4 x2 + 8 x3 = 0
The matrix operations are shown in Table 2.19.
The characteristic polynomial is found from the trace values.

( 1)3 (l 3 9l 2 ( 26) l 24) = 0


l 3 + 9l 2 26l + 24 = 0
The solution to the cubic equation can be found by many of the methods
from Chapter 1 and represent the eigenvalues l = 2, 3, and 4.

2.15 POWER METHOD OR ITERATION METHOD


The power method is an iterative method used when only the smallest or
largest eigenvalues and eigenvectors are desired. It may also be used to
find intermediate eigenvalues and eigenvectors using a sweeping technique. The sweeping technique can be found in Applied Numerical
Methods for Digital Computations, By M.L. James, G.M. Smith, and
J.C. Wolford. The largest eigenvalue is found by iterating on the equation
[A][x]= l[x].
The steps of procedure are as follows:
1. Assume values for the components of the eigenvector [x]=1.
2. Multiply the coefficient matrix times the vector [A][x].
3. Normalize the right hand side of the equation as follows l[x]:

92NUMERICAL STRUCTURAL ANALYSIS

Table 2.19. Example 2.16 FaddevLeverrier method


[B1]=[A] =

1
10
2

2
0
4

3
2
8

[B2]=([A]([B1]p1I)
1
10
2

2
0
4

3
2
8

1
10
2

2
0
4

3
2
8

[B3]=([A]([B2]p2I)
1
10
2

2
0
4

3
2
8

1
10
2

2
0
4

3
2
8

1
10
2

2
0
4

3
2
8

p1=tr[B1] = 9
1
10
2

2
0
4

3
2
8

8
10
2

2
9
4

3
2
1

9
9
9
34 4
4
76 12 32
40 8 6

p2=(1/2)tr[B2] = 26
34 4
4
76 12 32
40 8 6

4
32
20

8
76
40

4
14
8

26
26
26
24
0
0

0
24
0

0
0
24

p3=(1/3)tr[B3] = 24

a. Divide all the xs by first x value.


b. Divide all the xs by the largest x.
c. Normalize to a unit length.
4. Use the components of the normalized vector as improved values
of x.
5. Repeat steps 2 through 4 until the previous values differ from the
new values by less than some small value (e).

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 93

The smallest eigenvalue is found by iterating on the equation [A1][x]=


l1[x] in the same manner as the largest value. A common structural problem is the modal node analysis of a multi-story frame. The general steps
are as follows:
[K][x] [M][x] = 0
The value l = w2, where w is the frequency of the building. The equation
can be rewritten as follows:
1

[ B ][ M ][ x ] = l [ x ]
The values of the matrices are [K] for stiffness, [M] for mass, and [B] for
flexibility.
Example 2.17 Power method
Determine the first mode shape (lowest eigenvector) for the following set
of equations using the power method with e = 0.001 and assuming [x] =
[1] as an initial value. The [B] and [M] matrices are for a four-story single
mass structural model.
106

1 / 12
1 / 12
1 / 12 1 / 12
1 / 12 11 / 60 11 / 60 11 / 60

1 / 12 11 / 60 37 / 120 37 / 120

1 / 12 11 / 60 37 / 120 57 / 120

6
0

0
5
0
0

0
0
4
0

103

0
0
0

0
0
=
0

0

The [B] and [M] matrices can be combined by matrix multiplication.


0.000500
0.000500

0.000500

0.000500

0.000417
0.000917
0.000917
0.000917

0.000333
0.000733
0.001233
0.001233

0.000250 0
0.0000550 0
=
0.000925 0

0.001425 0

The iteration process and the final solution are shown in Table 2.20. The
xvector was normalized to the top value.

94NUMERICAL STRUCTURAL ANALYSIS

Table 2.20. Example 2.17 Power method


0.0005

0.0004167

0.0003333

0.00025

0.0005

0.0009167

0.0007333

0.00055

0.0005

0.0009167

0.0012333

0.000925

0.0005

0.0009167

0.0012333

0.001425

1
0.0015

0.0027

1.8
2.3833333

0.004075

2.7166667

1
1.9797042

0.007602

2.79128

0.008961

3.2900051

1
2.0050559

0.008801

2.8593634

0.010446

3.3938361

4
=

0.006301

2.0087357

0.009004

2.8701807

0.010701

3.4111138

5
0.003146

emax= 0.5733384

0.0030778

emax= 0.103831

0.003137

emax = 0.0172777

0.0031465

emax = 0.002883

5
=

0.006322

2.0093106

0.009037

2.8719442

0.010742

3.4139968

6
0.003148

0.0027236

3
=

0.006171

0.003137

0.0015

2
=

0.005392

0.003078

1
=

0.003575

0.002724

6
=

0.003148

emax = 0.000484

0.006326

2.0094044

e = 0.001

0.009042

2.87223733

l1= 317.6593

0.010749

3.4144812

SOLUTIONS OF SIMULTANEOUS LINEAR ALGEBRAIC EQUATIONS 95

REFERENCES
Anon 179. The Nine Chapters on the Mathematical Art. China.
Clasen, B.I. 1888. Sur une nouvelle method de resolution de equations lineaires et
sul lapplication de cette method au calcil des determinates. 3rd ed. Brussels,
Belgium.
Commandant Benoit. 1924. Note sur une mthode de rsolution des quations
normales provenant de lapplication de la mthode des moindres carrs
un systme dquations linaires en nombre infrieur celui des inconnues,
Bulletin Godsique. Heidelberg, Germany.
Cramer, G. 1750. Introduction lanalyse des lignes courbes algbriques. Geneva,
Switzerland.
Gauss, C.F. 1903. Werke 9, Gttingen, Germany.
Le Verrier, U. 1839. Thesis on the secular variations of the orbits of the planets.
Paris, France: Academy of Sciences.
Newton, I. 1707. Arithmetica Universalis. London, England.

CHAPTER 3

Numerical Integration and


Differentiation
The integration of a continuous function is used to find the area under the
function and to evaluate integral relationships of functions. Differentiation evaluates the rate of change of one variable with respect to another.
Examples of structural engineering problems involving integration and
differentiation include geometrical properties of centroids of areas and
volumes; moment of inertia; relationships between load, shear, moment,
rotation, and deflection of beams using the equation of the elastic curve;
and other strain energy relationships of structures involving shear, torsion,
and axial forces. Many methods exist to solve such types of problems with
varying levels of exactness. These and other problems will be covered in
the following chapters.

3.1TRAPEZOIDAL RULE
Consider a function f(x) graphed between points a and b along the x-axis
as shown in Figure3.1. One approximation of the area under the curve is
to apply the trapezoidal rule by dividing the area into n strips of width Dx.
Then, approximate the area of each strip as a trapezoid.
Calling the ordinates f(xi) = yi (i = 1, 2, 3,, n, n+1), the areas of each
strip are as follows:
y + y3
y + y4
y + y2
A1 = x 1
, A = x 2
, A = x 3
, An
2 2
2 3
2
y + y n +1
= x n

2
b

A = f ( x ) dx = A1 + A2 + A3 +  + An
a

A = f ( x ) dx =
a

A=

y +

x
( y1 + 2 y2 + 2 y3 +  + 2 yn + yn +1 )
2
2y + y

98NUMERICAL STRUCTURAL ANALYSIS


y=f(x)
y4

yn1

y3

yn
yn+1

y2

y1
A1

A2

A3

x=a

An1

An

xi

x=b

y + y3
y + y4
y + y2
A1 = x 1
, A2 = x 2
, A3 = x 3

, An
2 x x
2
x
x x 2
y + yn +1rule.

Figure3.1.
x n
= Trapezoidal

A = f ( x ) dx = A1 + A2 + A3 +  + An
a

A = f ( x ) dx =
a

A=

x
( y1 + 2 y2 + 2 y3 +  + 2 yn + yn +1 )
2

x
+
y
2 yi + yn +1

2
i=2

Example 3.1 Trapezoidal rule


Determine the area under the curve from 0 to p for y=sin(x) using the
trapezoidal rule with 2 and 4 strips.
A=

x
2 yi + yn +1
y
+

2
i=2

Two strips are shown in Table3.1:


Table3.1. Example 3.1 Trapezoidal rule
x

p/2

1.5708

3.1416

y=sin(x)

1.23E-16

A=

p
p
0 + 2 (1) + 0) = = 1.5708
(
4
2

Numerical Integration and Differentiation 99

Four strips are shown in Table3.2:


Table3.2. Example 3.1 Trapezoidal rule
x

p/4

p/2

3p/4

0.7854

1.5708

2.3562

3.1416

y=sin(x)

0.7071

0.7071

1.23E-16

A=

p
(0 + 2 (0.7071) + 2 (1) + 2 (0.7071) + 0) = p8 ( 4.8284) = 1.8961
8

The exact solution may be found by the integral:


p

sin ( x) dx = cos ( x) |

p
0

= cos ( p ) cos ( 0) = 1 + 1 = 2

3.2 ROMBERG INTEGRATION


A more accurate integral can be obtained using Rombergs method
(Romberg 1955). If a function can be defined as a continuous mathe
matical expression having continuous derivative f(x) and f(x), the
error of the trapezoidal rule is shown in Figure3.2 and can be found
as follows:
Expanding yi+1 in a Taylor series about xi and letting Dx=h as
follows:
yi +1 = yi + yi h +

yih 2 yih3
+
+ Higher order terms
2!
3!

The change in y between points iand i+1 is equal to the area under the
ycurve between those two points, therefore the exact area in the strip is
as follows:
yih 2 yih3
+
+ Higher order terms
2!
3!
f h 2 f h3
yi +1 yi = f i h + i + i
+ Higher order terms (3.1)
2!
3!

yi +1 yi = yih +

100NUMERICAL STRUCTURAL ANALYSIS

y = f(x)

fi

fi+1
x

x
y

yi+1 yi

yi+1
yi

xi

xi+1

Figure3.2. Romberg integration.

The Taylor series for fi+1 expanded about xi is as follows:

f h 2
f i +1 = f i + f i h + i
+ Higher order terms
2!
f i h = f i +1 f i

f i h 2
Higher order terms
2!

(3.2)

Substituting Equation 3.2 into Equation 3.1, the following can be derived:

f h 2
f i +1 f i i
Higher order terms h
2!

yi +1 yi = f i h +
2!
+

f i h3
+ Higher order terms
3!

yi +1 yi = exact area =

( fi + fi +1 ) h
2

f i h3
+ Higher order terms (3.3)
12

Numerical Integration and Differentiation 101

The first term on the right side in Equation 3.3 is the area or the trapezoid
and the rest is the error as follows:
ET =

f i h3
+ Higher order terms
12

The exact integral, I,can be derived using this error relation from two separate approximate integrals. The derivation is omitted from this text, but
may be found in Applied Numerical Methods for Digital Computations,
by James, Smith, and Wolford (1977). The improved integral is based on
two approximate integrals with a strip where h2 < h1 as follows:
I I
I I
I I h 2 + h12 h22 h22 = I h 2 + h1 2h 2
h1 1
h2 h1
h2

Ih2

h1


h2 I h1
2

h1 1
h2

If the second integration uses a strip one-half that of the first with h1/h2=2,
the equation becomes the following:
I h 2 ( 2) I h1
2

( 2) 2 1

This is a defined as a first-order extrapolation. If two first-order extrapolations are performed, then their results can be combined into a secondorder relationship with the following:
I h 2 ( 2) I h1
4

( 2) 4 1

The general nth order extrapolation would take the following form with n
being the order of extrapolation:
I

I h 2 4n I h1
4n 1

102NUMERICAL STRUCTURAL ANALYSIS

Example 3.2 Romberg integration


Determine the third-order extrapolation for the area under the curve from
0 to 1 for y=10x using Romberg integration along with the trapezoidal rule.
A=

x
y1 + 2 yi + yn +1

2
i=2

The trapezoidal integration for one strip is as follows in Table3.3:


Table3.3. Example 3.2 Romberg integration
x
y

0
1.0000
A=

1.000
10.0000

1
(1 + 10) = 5.5
2

Two strips are shown in Table3.4:


Table3.4. Example 3.2 Romberg integration
x
y
A=

0
1.0000

0.500
3.1623

1.000
10.0000

1
(1 + 2 (3.1623) + 10) = 4.33114
4

Four strips are shown in Table3.5:


Table3.5. Example 3.2 Romberg integration
x
y

0
1.0000

A=

0.250
1.7783

0.500
3.1623

0.750
5.6234

1.000
10.0000

1
(1 + 2 (1.7783) + 2 (3.1623) + 2 (5.6234) + 10) = 4.01599
8

Eight strips are shown in Table3.6:


Table3.6. Example 3.2 Romberg integration
x
y

0
1.0000

0.125 0.250 0.375 0.500 0.625 0.750 0.875


1.000
1.3335 1.7783 2.3714 3.1623 4.2170 5.6234 7.4989 10.0000

Numerical Integration and Differentiation 103

A=

1
(1 + 2 (1.3335) + 2 (1.7783) + 2 ( 2.3724) + 2 (3.1623)
16
+ 2 ( 4.2170) + 2 (5.6234) + 2 ( 7.4989) + 10) = 3.93560

From Romberg integration,


I

I h 2 4n I h1
4n 1

First order:
4.33114 ( 4) 5.5
1

( 4)1 1

= 3.941518

4.01599 ( 4) 4.33114
1

( 4)1 1

3.93560 ( 4) 4.01599

= 3.910944

( 4)1 1

= 3.908798

Second order:
3.910944 ( 4) 3.941518
2

( 4) 2 1
2
3.908798 ( 4) 3.910943
I
( 4) 2 1

= 3.9089055
= 3.9808648

Third order:
3.9086548 ( 4) 3.9089055
3

5.5

( 4 )3 1

4.33114
3.941518

= 3.98086509

4.01599
3.910944

3.9089055

3.908798
3.6086548

3.98086509

3.9560

104NUMERICAL STRUCTURAL ANALYSIS

The exact solution may be found from the following integral:


1

10
0

dx =

10 x 1
|0 = 4.342944819 0.434294482 = 3.9808650337
ln 10

3.3 SIMPSONS RULE


More accurate integration can be achieved by Simpsons rules credited
to Simpson (1750). Consider a function f(x) graphed between x=Dx and
x=Dx as shown in Figure3.3. An approximation of the area under the
curve between these two points would be to pass a parabola through the
points and zero (three points). The general second-degree parabola connecting the three points is as follows:
y = f ( x ) = ax 2 + bx + c
A=

A=

xx

ax 3 bx 2
+
+ cx
ax 2 + bx + c dx =
3
2
x

2
3
a ( x ) + 2c ( x )
3

(3.4)

The constants a, b, and c are found using the three points (Dx, yi), (0, yi+1),
and (Dx, yi+2) as follows:

y=f(x)

(xi, yi)

(xi+1, yi+1)

(xi+2, yi+2)
y=ax2+bx+c
x

x
Figure3.3. Simpsons rule.

Numerical Integration and Differentiation 105

yi = a ( x ) + b ( x ) + c
2

yi +1 = c
yi + 2 = f ( x) = a (x) 2 + b(x) + c
Solving the three equations with three unknowns we obtain the following:
a=
b=

yi 2 yi +1 + yi + 2
2 ( x)

yi + yi + 2
2 ( x )

c = yi +1
Substituting the expressions for a and c into Equation 3.4, the following
is achieved:
A=

x
( yi + 4 yi +1 + yi + 2 )
3

If we apply this to n even numbered strips, the following occurs:


x
( y1 + 4 y2 + y3 )
3
x
A=
( y3 + 4 y4 + y5 )
3
x
A=
( y5 + 4 y6 + y7 )
3
x
A=
( yn 1 + 4yn + yn +1 )
3
A=

In general form, this is the following:


A=

n 1
n

x
y1 + 4 yi + 2 yi + yn +1

i = 3, 5 , 7
i = 2, 4,6

If we performed the error truncation to obtain Rombergs integration, the


following occurs:
I

I h 2 16n I h1
16n 1

106NUMERICAL STRUCTURAL ANALYSIS

Similarly, Simpsons three-eighths rule can be derived using three strips


and a third-degree parabola. The following is the solution:
y = f ( x ) = ax 3 + bx 2 + cx + d
A=

3 x
2

3 x

3 x
2

A=

2
ax 4 bx 3 cx 2
+ dx
+
+
ax + bx + cx + d dx =
3
2
3x
4
3

9
3
b ( x ) + 3d ( x )
4

(3.5)

The constants a, b, c, and d are found using the four points (3Dx/2, yi),
(Dx/2, yi+1), (Dx/2, yi+2), and (3Dx/2, yi+3) as follows:
3

3 x
3 x
3 x
yi = a
+ b
+ c
+d

2
2
2
3

x
x
x
yi +1 = a
+ b
+ c
+d
2
2
2
2

x
x
x
yi + 2 = a + b + c + d
2
2
2
3

yi + 3

3 x
3 x
3 x
+d
= a
+ b
+ c
2
2
2

Solve the four equations with four unknowns and then substitute these
back into Equation 3.5 to achieve the following:
A=

3 x
( yi + 3 yi +1 + 3 yi + 2 + yi + 3 )
8

The general form with n strips is as follows:


A=

n 1, n
n2

3 x
3
2
y
+
y
+
yi + yn +1

1
i

i = 2 , 3, 5 , 6
i = 4, 7

For an odd number of strips, both the one-third and three-eighths rules
must be used. The three-eighths rule is used to obtain the area contained
in three strips under the curve and then the one-third rule is used for the
remaining n3 strips.

Numerical Integration and Differentiation 107

Example 3.3. Simpsons one-third rule


Determine the area under the curve from p/2 to p/2 for y = x2cos(x) using
Simpsons one-third rule with four and eight strips.

A=

n 1
n

x
4
2
y
+
y
+
yi + yn +1

i
1

i = 3, 5 , 7
i = 2, 4,6

Four strips are shown in Table3.7:


Table3.7. Example 3.3 Simpsons one-third rule
-p/2
-1.57080
0.00000

x
y

-p/4
-0.78540
0.43618

0
0.00000
0.00000

p/4

p/2

0.78540
0.43618

1.57080
0.00000

p
p
0 + 4 ( 0.43618) + 2 ( 0) + 4 ( 0.43618) + 0) = (3.48944)
(
12
12
= 0.913533

A=

Eight strips are shown in Table3.8:


Table3.8. Example 3.3 Simpsons one-third rule
-p/2

-3p/8

-p/4

-p/8

p/8

p/4

3p/8

p/2

x -1.57080 -1.17810 -0.78540 -0.39270 0.00000 0.39270 0.78540 1.17810 1.57080


y 0.00000

A=

0.53113

0.43618 0.14247 0.00000 0.14247 0.43618 0.53113 0.00000

p
(0 + 4 (0.53113) + 2 (0.43618) + 4 (0.14247) + 2 (0) + 4 (0.14247)
24
p
+ 2 ( 0.43618) + 4 ( 0.53113) + 0) =
(7.13352) = 0.933776
24

Perform Romberg extrapolation with these two integrations to get a more


exact solution as follows:

I h 2 16n I h1 ( 0.933776)161 0.913533


=
= 0.935126
16n 1
161 1

108NUMERICAL STRUCTURAL ANALYSIS

The exact solution may be found by the integral:


p

x cos ( x) dx = 2xcos ( x) + ( x

2
2 sin ( x ) p

= ( 0 + 0.4674011003) ( 0 0.4674011003) = 0.9348022005

Example 3.4. Simpsons one-third and three-eighths rules


Determine the area under the curve from 0 to p/2 for y = sin3x+cos3x using
Simpsons three-eighths and one-third rules (in that order) with five strips.
Simpsons three-eighths rule is set up in Table3.9:
A=

3 x
( yi + 3 yi +1 + 3 yi + 2 + yi + 3 )
8

Table3.9. Example 3.4 Simpsons one-third and three-eighths rules


0
p/10
p/5
3p/10
2p/5
p/2
0.00000 0.31416 0.62832 0.94248 1.25664 1.57080
1.00000 0.88975 0.73258 0.73258 0.88975 1.00000

x
y
A=

3p
3p
1 + 3 ( 0.88975) + 3 ( 0.73258) + 0.73258) =
(6.59957) = 0.777494
(
80
80

Simpsons one-third rule is set up in Table3.10:


x
A=
( yi + 4 yi +1 + yi + 2 )
3
Table3.10. Example 3.4 Simpsons one-third and three-eighths rules
0
0.00000
1.00000

A=

p/10

p/5

3p/10

0.31416
0.88975

0.62832
0.73258

0.94248
0.73258

2p/5
1.25664
0.88975

p/2

1.57080
1.00000

p
p
0.73258 + 4 ( 0.88975) + 1) = (5.29158) = 0.554133
(
30
30

Adding the two together for a total area:


A = 0.777494 + 0.554133 = 1.331627

Numerical Integration and Differentiation 109

The exact solution may be found by the integral:


p

( sin x + cos x ) dx = ( sin


2

2 n 1

x + cos 2 n 1 x dx

2 ( 4) 6 ( 2n)
2 ( 4) 6 ( 2n)
2 2 4
+
= + =
3 (5) 7 ( 2n + 1) 3 (5) 7 ( 2n + 1) 3 3 3

3.4GAUSSIAN QUADRATURE
The main difference of Gaussian quadrature from the previous methods
is that the interval to be integrated is not divided into strips. Instead, a
central point is used to determine the best places to evaluate the function.
The Gauss points indicate how far from the central point to go and then
each point is weighted. The derivation of this method is not included
here, but can be found in many advanced mathematics textbooks. The
method is named for Carl Friedrich Gauss (1801). The following is a general equation that shows the process for n Gauss points for integration:
b

i =1

A = f ( x ) dx = s f ( s1 sxi ) wi
ba
2
b+a
s1 =
2
s=

The number of points used should closely match the degree of the equation
to integrate. Table3.11 shows some of the Gaussian quadrature points, xi,
and their weights, wi.
Example 3.5. Gaussian quadrature
Determine the area under the curve from 1 to 10 for y = log10x using
Gaussian quadrature with 2, 3, and 4 points.
b

i =1

A = f ( x ) dx = s f ( s1 sxi ) wi
b a 10 1
s=
=
= 4.5
2
2
b + a 10 + 1
s1 =
=
= 5.5
2
2

110NUMERICAL STRUCTURAL ANALYSIS

Table3.11. Gaussian quadrature


xi
0
-0.577350269
0.577350269
-0.774596669
0
0.774596669
-0.861136312
-0.339981044
0.339981044
0.861136312
-0.906179846
-0.538469310
0
0.538469310
0.906179846
-0.932469514
-0.661209386
-0.238619186
0.238619186
0.661209386
0.932469514

One point
Two points
Three points

Four points

Five points

Six points

wi
2
1
1
0.555555556
0.888888889
0.555555556
0.347854845
0.652145155
0.652145155
0.347854845
0.236926885
0.478628670
0.568888889
0.478628670
0.236926885
0.171324492
0.360761573
0.467913935
0.467913935
0.360761573
0.171324492

Two points are shown in Table3.12:


Table3.12. Example 3.5 Gaussian quadrature
xi

-0.577350269

2-points

0.577350269

wi

s1+sxi

2.90192379

f(s1+sxi)

0.462686003

f(s1+sxi)wi

0.462686003

1
8.098076211
0.90838186
0.90838186
S

1.371067862

S*s

6.169805381

Numerical Integration and Differentiation 111

Three points are shown in Table3.13:


Table3.13. Example 3.5 Gaussian quadrature
xi
3-point

-0.774596669
0
0.774596669
0.555555556
0.888888889
0.555555556
2.01431499
5.5
8.985685011
0.304127385
0.740362689
0.953551191
0.168959658
0.658100169
0.529750662
1.356810489

S*s

6.105647198

wi

s1+sxi

f(s1+sxi)

f(s1+sxi)wi

Four points are shown in Table3.14:


Table3.14. Example 3.5 Gaussian quadrature
xi

-0.861136312

4-points

-0.339981044
0.339981044
0.861136312
0.347854845
0.652145155
0.652145155
0.347854845
1.624886596
3.970085302
7.029914698
9.375113404
0.210823056
0.598799838
0.846950055
0.97197653
0.073335822
0.390504413
0.552334375
0.338106745
1.354281355

S*s

6.094266097

wi

s1+sxi

f(s1+sxi)

f(s1+sxi)wi

112NUMERICAL STRUCTURAL ANALYSIS

The exact solution may be found by the integral:


10

log ( x) dx = [ xlogx xloge]


1

10
1

= 5.65706 + 0.43429 = 6.09135

3.5DOUBLE INTEGRATION BY SIMPSONS


ONE-THIRD RULE
When using double integration by Simpsons one-third rule, weighting
is applied in both directions and is then multiplied by the spacing in both
directions. The following is a general weighting array for four strips:
1 4 2 4 1
4 16 8 16 4
2 8 4 8 2
4 16 8 16 4
1 4 2 4 1
For two strips, the weighting array is the following:
1 4 1
4 16 4
1 4 1
Any even set of strips will follow the same pattern and this could also
be done using any other type of integration. Using the trapezoidal rule
would be less accurate, but could do any number of strips and Simpsons
three-eighths rule would require a multiple of three strips in each direction. The summation of the weighting array multiplied by f(x,y) is used in
the following equation to obtain the volume.
hxhy
= f ( x, y ) = V
9
The terms hx and hy are the spacing in the x and y directions, respectively.
Example 3.6 Double integration by Simpsons rule
Determine the volume under the hyperbolic paraboloid from x=0 to 8 and
y=0 to 8 for 0 = 16z xy using Simpsons one-third rule with four strips
in each direction.

Numerical Integration and Differentiation 113

x=0 to 8 Dx=2=hxand
y=0 to 8 Dy=2=hy
hxhy
= f ( x, y ) = V
9
Weighting operator:
1 4 2 4 1
4 16 8 16 4
2 8 4 8 2
4 16 8 16 4
1 4 2 4 1
Solving for z:
z=

xy
16

Table3.15 shows the set up and summation as follows:


hxhy
= f ( x, y ) = V
9
2 ( 2)
144 = 64 = V
9
Table3.15. Example 3.6 Double integration by Simpsons
one-third rule
x
0
0
0
0
0
2
2
2
2

y
0
2
4
6
8
0
2
4
6

weight
1
4
2
4
1
4
16
8
16

f(x,y)
0
0
0
0
0
0
0.25
0.5
0.75

weight*f(x,y)
0
0
0
0
0
0
4
4
12
(Continued)

114NUMERICAL STRUCTURAL ANALYSIS

Table3.15.(Continued)
x
2
4
4
4
4
4
6
6
6
6
6
8
8
8
8
8

y
8
0
2
4
6
8
0
2
4
6
8
0
2
4
6
8

weight
4
2
8
4
8
2
4
16
8
16
4
1
4
2
4
1

f(x,y)
1
0
0.5
1
1.5
2
0
0.75
1.5
2.25
3
0
1
2
3
4
S

weight*f(x,y)
4
0
4
4
12
4
0
12
12
36
12
0
4
4
12
4
144

3.6DOUBLE INTEGRATION BY GAUSSIAN


QUADRATURE
Double integration by Gaussian quadrature is very similar to the single
integration process. In this case, the Gauss equation is applied in both
directions and then multiplied by the weighting factors.
V =

bx , by
ax , bx

f ( x, y ) dxdy = sx s y f sx1 sx xi , s y1 sx yi wxi wyi


i =1

bx ax
b + ax
and sx1 = x
2
2
by a y
by + a y
sy =
and s y1 =
2
2
sx =

The number of points used should closely match the degree of the equation to integrate. The same Gauss points and weights from Section 3.4 are
used in each direction.

Numerical Integration and Differentiation 115

Example 3.7 Double integration by Gaussian quadrature


Determine the volume under the hemisphere from x=4 to 4 and y=4
to 4 for 64=x2+y2+z2 using Gaussian quadrature with three points in each
direction.
Solving for z and the points:
z = 64 x 2 y 2
bx ax 4 ( 4)
=4
=
2
2
b + ax 4 + ( 4)
=0
=
sx1 = x
2
2
by a y 4 ( 4)
=
=4
sy =
2
2
by + a y 4 + ( 4)
s y1 =
=0
=
2
2
sx =

Table3.16 shows the set up and summation as follows with xi, yi given
values and wx, wy corresponding weights when using three points:
Table3.16. Example 3.7 Double integration by Gaussian quadrature
xi

yi

x=

y=

sx1 + sxxi

sy1 + svyi

wx

wy

f(x,y)

wx*wy*f(x,y)

-0.774597 -0.774597 -3.098387 -3.098387 0.555556 0.555556 6.69328

2.065827

-0.774597

0.555556 0.888889 7.375636

3.642289

-0.774597 0.774597 -3.098387 3.098387 0.555556 0.555556 6.69328

2.065827

-3.098387

-0.774597

0.774597

-3.098387 0.888889 0.555556 7.375636


0

0.888889 0.888889

3.642289
6.320988

3.098387 0.888889 0.555556 7.375636

3.642289

0.774597 -0.774597 3.098387 -3.098387 0.555556 0.555556 6.69328

2.065827

0.774597

3.642289

0.774597

3.098387

0.555556 0.888889 7.375636

0.774597 3.098387 3.098387 0.555556 0.555556 6.69328


S

V = sx s y = 4 ( 4) 29.153 = 466.45

2.065827
29.153453

116NUMERICAL STRUCTURAL ANALYSIS

3.7TAYLOR SERIES POLYNOMIAL EXPANSION


The differentiation of a continuous function is used to find slopes, curvatures, and values for a function but can also be used to find other
relationships of functions. Often in structural engineering, there is a
need to find differential relationships. One simple way to easily evaluate transcendental equations is to use polynomial expansion developed
for the Taylor series. This is often referred to as the power series. The
general Taylor series polynomial expansion of a function is as follows:
y = f ( x ) = b0 + b1 x + b2 x 2 + b3 x 3 +  + bn x n
Successive derivatives of the function evaluated at zero can yield the coefficients, b.
f ( 0) = b0

f ( 0) = 1b1

f ( 0) = 1( 2) b2

f ( 0) = 1( 2) 3b3 = 3!b3
     
f

(0) = 1( 2) 3(i)bi = i !bi


     

(0) = 1( 2) 3(n)bn = n !bn

By taking successive derivatives of the function then evaluating them,


the coefficients of the polynomial may be found. This is how most digital equipment like computers and calculators find values for transcendental
equations.
Example 3.8 Taylor series polynomial expansion
Expand y=sin(x) into a polynomial using Taylor series including up to the
ninth degree term. Check by calculating sin45.
y = sin ( x ) = b0 + b1 x + b2 x 2 + b3 x 3 +  + b9 x 9

Numerical Integration and Differentiation 117

Evaluate the successive derivatives as follows:


f ( 0) = b0 = sin ( 0) = 0 b0 = 0

f ( 0) = 1b1 = cos ( 0) = 1 b1 = 1

f ( 0) = 2b2 = sinn ( 0) = 0 b2 = 0
f ( 0) = 3!b3 = cos ( 0) = 1 b3 =
f ( 0) = 4 !b4 = sin ( 0) = 0 b4 = 0

1
3!

1
5!
f ( 0) = 6 !b6 = sin ( 0) = 0 b6 = 0
f ( 0) = 5!b5 = cos ( 0) = 1 b5 =

f ( 0) = 7 !b7 = cos ( 0) = 1 b7 =
f ( 0) = 8!b8 = sin ( 0) = 0 b8 = 0
f ( 0) = 9 !b9 = cos ( 0) = 1 b9 =

1
7!

1
9!

The polynomial can then be written with the coefficients.


y = sin ( x ) = x

1 3 1 5 1 7 1 9
x + x x + x
3!
5!
7!
9!

The value of sin45=sin(p/4) can be evaluated to check the accuracy of


the approximation.

( )

( )

( )

( )

( )

3
5
1
1
1 p 7 1 p 9
y = sin p 4 = p 4 p 4 + p 4
+
3!
5!
7! 4
9! 4
y = 0.7853981634 0.0808455122 + 0.0024903946 0.0000365762
+ 0.0000003134
y = 0.7071067829

The exact solution to the same accuracy is y=0.7071067812.


Example 3.9. Taylor series polynomial expansion
Expand y=ex into a polynomial using Taylor series including up to the
fifth degree term. Check by calculating e1.
y = e x = b0 + b1 x + b2 x 2 + b3 x 3 +  + b5 x 5

118NUMERICAL STRUCTURAL ANALYSIS

Evaluate the successive derivatives as follows:


f ( 0) = b0 = e 0 = 1 b0 = 1
f (0) = b1 = e 0 = 1 b1 = 1
f (0) = 2b2 = e 0 = 1 b2 =

1
2

f (0) = 3!b3 = e 0 = 1 b3 =
f (0) = 4!b4 = e 0 = 1 b4 =

1
4!

1
3!

f (0) = 5!b5 = e 0 = 1 b5 =

1
5!

The polynomial can then be written with the coefficients.


y = e x = 1 x +

1 2 1 3 1 4 1 5
x x + x x
2!
3!
4!
5!

The value of the e1 can be evaluated to check the accuracy of the


approximation.
1 2 1 3 1 4 1 5
(1) (1) + (1) (1)
2!
3!
4!
5!
y = 1 1 + 0.5 0.166667 + 0.041667 0.008333
y = 0.366667
y = e x = 1 1 +

The exact solution to the same accuracy is y = 0.367879. If the polynomial


was calculated up to the x9 term, the value would be y = 0.3678791 versus
the exact value of y = 0.3678794.

3.8DIFFERENCE OPERATORS BY TAYLOR


SERIES EXPANSION
The numerical differential equation relationships can be found using the
Taylor series expansion. Expanding the Taylor series for a function y=f(x)
at x=(xi+h) gives the following equation:
y ( xi + h ) = yi + yi h +

yih 2 yih3
+
+
2!
3!

(3.6)

Numerical Integration and Differentiation 119

Next, expanding the Taylor series for a function y=f(x) at x=(xih) gives
the following equation:
y ( xi h ) = yi yi h +

yi' h 2
2!

yih3
+
3!

(3.7)

If Equation 3.7 is subtracted from Equation 3.6, a first derivative relationship is as follows:
yi =

y ( xi + h ) y ( xi h ) yih 2

+ 
2h
6

(3.8)

This may be written as follows with successive values of y and the higher
order terms omitted:
yi =

yi +1 yi 1
2h

This equation is known as the central-difference approximation of yi at xi


with errors, order of h2. If Equation 3.7 is added to Equation 3.6, a second
derivative relationship is as follows:
yi=

y ( xi + h ) 2 yi + y ( xi h ) yih 2

+  (3.9)
2
h
12

This may be written as follows with successive values of y and the higher
order terms omitted:
yi=

yi +1 2 yi + yi 1
h2

This equation is known as the central-difference approximation of yi at xi


with errors, order of h2. Next, expanding the Taylor series for a function
y=f(x) at x=(xi+2h) gives the following equation:
y ( xi + 2h ) = yi + yi 2h +

yi ( 2h)
y ( 2h) yi( 2h)
+ i
+
+  (3.10)
2!
3!
4!
2

120NUMERICAL STRUCTURAL ANALYSIS

Finally, expanding the Taylor series for a function y=f(x) at x=(xi2h)


gives the following equation:
y ( xi 2h ) = yi yi2h +

yi ( 2h)
y ( 2h)
y ( 2h)
i
+ i
 (3.11)
2!
3!
4!
2

If Equation 3.11 is subtracted from Equation 3.10 and the equation for the
first derivative is substituted into the result, a third derivative relationship
is as follows, with an order of error h2:
yi=

yi + 2 2 yi +1 + 2 yi 1 yi 2
2h 3

If Equation 3.11 is added to Equation 3.10 and the equation for the second
derivative is substituted into the result, a fourth derivative relationship is
as follows with an order of error h2:
yi =

yi + 2 4 yi +1 + 6 yi 4 yi 1 + yi 2
h4

These are the central difference expressions with error order h2. Higher
order expressions can be derived if we include more terms in each expansion. Forward difference expressions can be derived by using Taylor series
expansion of x=(xi+h), x=(xi+2h), x=(xi+3h), and so forth. Backward
difference expressions may also be derived by Taylor series expansion of
x=(xih), x=(xi2h), x=(xi3h), and so forth. The following are the central, forward, and backward difference expressions of varying error order.
These were compiled from Applied Numerical Methods for Digital Computations, by James, Smith, and Wolford (1977). They can also be written
in a reverse graphical form that is sometimes used and compiled from
Numerical Methods in Engineering, by Salvadori and Baron (1961).
Central difference expressions with error order h2:
yi +1 yi 1
2h
yi +1 2 yi + yi 1
yi=
h2
y 2 yi +1 + 2 yi 1 yi 2
yi = i + 2
2h 3
y 4 yi +1 + 6 yi 4 yi 1 + yi 2
yi= i + 2
h4
yi =

Numerical Integration and Differentiation 121

2hD
h2 D 2
2h3 D 3
h4 D 4

i 2 i 1 i i +1 i + 2
1 0
1
1 2 1
1
2
0 2
1
1
4 6 4
1

Central difference expressions with error order h4:


yi + 2 + 8 yi +1 8 yi 1 + yi 2
12h
y + 16 yi +1 30 yi + 16 yi 1 yi 2
yi = i + 2
12h 2
y + 8 yi + 2 13 yi +1 + 13 yi 1 8 yi 2 + yi 3
yi = i + 3
8h 3
y + 12 yi + 2 39 yi +1 + 56 yi 39 yi 1 + 12 yi 2 yi 3
yi = i + 3
6h 4
yi =

12hD
12h 2 D 2
8h3 D 3
6h 4 D 4

i 3 i 2 i 1 i
i +1 i + 2 i + 3
1
8
0
8
1
1 16 30 16
1
1
8 13
0 13
8
1
1 12 39 56 39 12
1

Forward difference expressions with error order h:


yi +1 yi
h
y
2 yi +1 + yi
+
2
i
yi =
h2
y 3 yi + 2 + 3 yi +1 yi
yi = i + 3
h3
y 4 yi + 3 + 6 yi + 2 4 yi +1 + yi
yi = i + 4
h4
yi =

hD
h2 D 2
h3 D 3
h4 D 4

i i +1 i + 2 i + 3 i + 4
1 1
1 2
1
1 3
3
1
1 4
6
4
1

122NUMERICAL STRUCTURAL ANALYSIS

Forward difference expressions with error order h2:


yi + 2 + 4 yi +1 3 yi
2h
yi + 3 + 4 yi + 2 5 yi +1 + 2 yi
yi =
h2
3 yi + 4 + 14 yi + 3 24 yi + 2 + 18 yi +1 5 yi
yi =
2h 3
2 yi + 5 + 11 yi + 4 24 yi + 3 + 26 yi + 2 14 yi +1 + 3 yi
yi=
h4
yi =

2hD
h2 D 2
2h 3 D 3
h4 D 4

i i +1 i + 2 i + 3 i + 4 i + 5
3 4
1
2 5
4
1
5 18 24 14
3
3 14 26 24 11
2

Backward difference expressions with error order h:


yi yi 1
h
yi 2 yi 1 + yi 2
yi =
h2
y 3 yi 1 + 3 yi 2 yi 3
yi = i
h3
y 4 yi 1 + 6 yi 2 4 yi 3 + yi 4
yi= i
h4
yi =

hD
h2 D 2
h3 D 3
h4 D 4

i 4 i 3 i 2 i 1
1
1
2
1
3
3
1
4
6
4

i
1
1
1
1

Backward difference expressions with error order h2:


3 yi 4 yi 1 + yi 2
2h
2 yi 5 yi 1 + 4 yi 2 yi 3
yi =
h2
5 y 18 yi 1 + 24 yi 2 14 yi 3 + 3 yi 4
yi = i
2h 3
3 y 14 yi 1 + 26 yi 2 24 yi 3 + 11 yi 4 2 yi 5
yi= i
h4
yi =

3 yi 4 yi 1 + yi 2
2h
2Numerical
yi 5 yi 1 + 4Integration
yi 2 yi 3
and Differentiation 123
yi =
h2
5 y 18 yi 1 + 24 yi 2 14 yi 3 + 3 yi 4
yi = i
2h 3
3 y 14 yi 1 + 26 yi 2 24 yi 3 + 11 yi 4 2 yi 5
yi= i
h4
yi =

2hD
h2 D 2
2h 3 D 3
h4 D 4

i 5 i 4 i 3 i 2 i 1
1
4
1
4
5
3
14 24 18
2
11 24 26 14

i
3
2
5
3

3.9NUMERIC MODELING WITH DIFFERENCE


OPERATORS
The difference operators, sometimes referred to as finite difference operators,
can be used to solve many structural engineering problems involving differential equation relationships. One common relationship is that of the equation of the elastic curve. The equation of the elastic curve relatesthe deflected
shape of a beam to the rotation, moment, shear, and load on thebeam. This is
a typical strength of materials topic and the following are the basic relationships based on the deflection equation in y and , M, V, and q:
yi = q
M
EI
V
yi =
EI
q
yi=
EI
yi =

Example 3.10 Simple beam with difference operator


Calculate the shear, moment, rotation, and deflection for a 25 foot long,
simply supported beam with a uniformly distributed load of 4 k/ft using
central difference operator of order of error h2 at 1/6th points. The beam has
E=40,000 ksi (modulus of elasticity) and I=1000 in4 (moment of inertia).
To solve the problem, a sketch of the beam and the assumed deflected
shape is created. To use central difference operators, the model must go
beyond the boundaries of the physical beam. The deflected shape must

124NUMERICAL STRUCTURAL ANALYSIS

also be modeled beyond those boundaries with some confidence. If the


general equation is continuous at the boundaries, then this type of model
is appropriate. If not, then the model should end at the boundaries and forward or central difference operators must be used. Figure3.4 shows the
beam and the assumed deflections at 1/6th points of the beam. By symmetry of the model, only four specific values of the deflection are unknown.
Z

4 k/ft
A

B
25-0

Y3

Y2 Y1

Y0

Y1

Y2

Y3

Y2

Y1

Y0

Y1 Y2 Y3

Figure3.4. Example 3.10 Simple beam with difference operator.

The central-difference expressions with error of order h2 will be used


to solve for the values. Since the load is known, we will use the fourth
derivative relationship between load and deflection.
y 4 yi +1 + 6 yi 4 yi 1 + yi 2
q
= i+2
EI
h4
EI
q = 4 ( yi + 2 4 yi +1 + 6 yi 4 yi 1 + yi 2 )
h
yi=

Placing the central difference operator on y0, the first equation can be written from Figure3.5:
Y3

Y2

Y1

Y0

Y1

Y2

Y3

Y2

Y1

Y0

Y1

Y2

Y3

Figure3.5. Example 3.10 Simple beam with difference operator.

q0 =

EI
EI
y2 ) 4 ( y1 ) + 6 y0 4 y1 + y2 ) = 4 (6 y0 )
4 ((
h
h

Numerical Integration and Differentiation 125

For the second equation, the central difference operator is placed on y1 and
is shown in Figure3.6.

Figure3.6. Example 3.10 Simple beam with difference operator.

q1 =

EI
(4 y0 + 5 y1 4 y2 + y3 )
(( y1 ) 4 y0 + 6 y1 4 y2 + y3 ) = EI
h4
h4

The third and fourth equations can be written by placing the central difference operator on y2 and y3.
EI
EI
y 4 y1 + 6 y2 4 y3 + y2 ) = 4 ( y0 4 y1 + 7 y2 4 y3 )
4 ( 0
h
h
EI
EI
q3 = 4 ( y1 4 y2 + 6 y3 4 y2 + y1 ) = 4 ( 2 y1 8 y2 + 6 y3 )
h
h
q2 =

These four equations constitute a non-homogeneous linear algebraic set


and can be written in matrix form.
6 0 0 0 y0 q0
4 5 4 1 y q
EI
1 = 1
4
h 1 4 7 4 y2 q2


0 2 8 6 y3 q3
From the conditions of the beam, two simplifications can be made. First,
the load is uniform and all the values of q are the same. Second, the deflection at point 0 is known to be zero, so the first row and column can be
eliminated since they correspond to those values.

5 4 1 y1 q
EI
4 7 4 y2 = q
h4
2 8 6 y3 q

This can be solved by many of the methods presented in Chapter 2. The


method of cofactors is used here, since the solution is small enough to
solve determinants directly.

126NUMERICAL STRUCTURAL ANALYSIS

2.50 4.00 2.25 1 y1


qh 4
4.00 7.00 4.00 1 = y2
EI
4.50 8.00 4.75 1 y3
When the value of h=L/6 is substituted in and the matrix multiplication
is performed, the following is a general solution for a simply supported
beam with a uniformly distributed load:
8.75 y1
qL4
15.00 = y2
1296 EI
17.25 y3
The particular solution for q=4 k/ft, L=25 ft, E=40,000 ksi, and I=1000 in4
is then obtained. The exact solution at the center is y3= 0.87891 inches,
which is a 2.22% error.
y1
0.44573
0.78125 inches = y
2

y3
0.89855
Once the deflections are found, the other three desired values, q, M, and
V, can each be found from the corresponding central difference operators
using the same order of error. The procedure is the same as the operator
is laid upon each of the values that are unknown then the corresponding
equation may be written. The relationship for rotation is as follows:
yi +1 yi 1
2h
0
2
0 0 y0 q 0

1 0 1 0 y q
1
1 = 1
2h 0 1 0 1 y2 q 2

0 0 0 0 y3 q 3
0 2 0 0 y0 q 0


3 1 0 1 0 y1 q1
=
L 0 1 0 1 y2 q 2

0 0 0 0 y3 q 3
2 y1 q 0

3 y0 + y 2 q 1
=
L y1 + y3 q 2

0
q 3

yi = q =

Numerical Integration and Differentiation 127

q0
0.00912
q
0.00781
radians = 1

q2
0.00443

0.0000
q3
The exact solution at the end is q0=0.00936 radians, which is a 2.51% error,
while the value in the center is exact. The relationship for moment is as follows:
y 2 yi yi 1
M
= i +1
EI
h2
2 0 0 0 y0 M 0


EI 1 2 1 0 y1 M 1
=
2
h 0 1 2 1 y2 M 2


0 0 2 2 y3 M 3

yi =

2 0 0 0 y0 M 0

36 EI 1 2 1 0 y1 M 1
2
=
L 0 1 2 1 y2 M 2

0 0 2 2 y3 M 3
2 y0

M 0
y 2y + y M
36 EI 0
1
1
2
2
=

+
y
y
y
M
2
L
2
1
2
3

2 y2 2 y3 M 3
M 0
0.00
M
2083.4

kip-inches = 1
M 2
3333.3

3750.0
M3
The exact solution at the center is M3=3750 k-in which is exact, while the
values at the ends is also exact. The relationship for shear is as follows:
y 2 yi +1 + 2 yi 1 yi 2
V
= i+2
EI
2h 3
0 4 2 0 y0 V0

EI 2 1 2 1 y1 V1
=
3
2h 1 2 1 2 y2 V2

0 0 0 0 y3 V3
yi =

0 4 2 0 y0 V0

108 EI 2 1 2 1 y1 V1
=

L3 1 2 1 2 y2 V2

y 2 yi +1 + 2 yi 1 yi 2
V
= i+2
EI
2h3
0 4 2 0 y0 V0

EI 2 1 2 1 y1 V1
=
3
1 2 y2 V2
h 1 2 ANALYSIS
128NUMERICAL 2STRUCTURAL

0 0 0 0 y3 V3
yi =

0 4 2 0 y0 V0

108 EI 2 1 2 1 y1 V1

=
3
L 1 2 1 2 y2 V2

0 0 0 0 y3 V3
4 y1 + 2 y2
V0

2 y + y 2 y + y V
108 EI 0
1
1
2
3

=
L3 y0 + 2 y1 + y2 2 y3 V2

0
V3

V0
35.27
V
34.92
kips = 1

V2
19.90

0
V3
The exact solution at the end is V0 = 50.00 k, which is a 16.67% error,
while the value in the center is exact. The large error at the end is due to
the fact that the shear drops there, which creates a discontinuity in the
equation. Using more segments would reduce the error, but it would still
be more inaccurate than the other values.
The next example is similar to the previous one, but is included to
show differences in modeling and accuracy. It also uses the higher order
(smaller error) of error equations for more accuracy.

Example 3.11 Fixed beam with difference operator


Calculate the shear, moment, rotation, and deflection for a 30 ft long fixed
end beam with a uniformly distributed load of 5 k/ft using central difference operator of order of error h4 at 1/6th points. The beam has E=29,000
ksi and I=1000 in4.
The primary difference in the simply supported beam in Example3.10
and the fixed end beam in this example is the model of the deflected curve
beyond the boundary as shown in Figure3.7.
The solution to this example is very similar to Example 3.10 and only
the setup and solutions are presented. The central difference expressions
with error of order h4 will be used to solve for the values. Since the load
is known, we will use the fourth derivative relationship between load and
deflection.

Numerical Integration and Differentiation 129


Z

5 k/ft
A

B
30-0

Y3

Y2

Y1

Y0

Y1

Y2

Y3

Y2

Y1

Y0

Y1

Y2

Y3

Figure3.7. Example 3.11 Fixed beam with difference operator.

yi=

q=

y + 12 yi + 2 39 yi +1 + 56 yi 39 yi 1 + 12 yi 2 yi 3
q
= i+3
EI
6h 4

EI
( yi + 3 + 12 yi + 2 39 yi +1 + 56 yi 39 yi 1 + 12 yi 2 yi 3 )
6h 4

Placing the central difference operator on all four unknown points in the
model, the linear non-homogeneous solution set can be obtained and
solved.
56 78 24 2 y0 q0

EI 39 68 41 12 y1 q1
4
=
6h 12 41 68 39 y2 q2

2 24 78 56 y3 q3
68
EI
4 41
6h
24
766
6qh 4
1360
EI
1566

41 12 y1 q
68 39 y2 = q
78 56 y3 q

1360 783
1 y1
1
3520 2160
1 = y2
15120
1 y3
4320 2943
2909 y1
q L4
7040 = y
2
3265920 EI
8829 y3

130NUMERICAL STRUCTURAL ANALYSIS

y1
0.2150
0.5202 inches = y
2

y3
0.6524
The exact solution at the center is y3=0.6284 inches, which is a 3.81%
error. This is more error than the deflection found in Example 3.10,
even though a more accurate operator was used. This is due to the
fact that this physical model has more variation in deflection than
that of the simply supported beam. The relationship for rotation is as
follows:
yi + 2 + 8 yi +1 8 yi 1 + yi 2
12h
0
0
0
0 y0 q 0

8 1 8 1 y q
1
1 = 1
12h 1 8 1 8 y2 q 2

0 0 0 0 y3 q 3
0 0 0 0 y0 q 0

1 8 1 8 1 y1 q1
=
2 L 1 8 1 8 y2 q 2

0 0 0 0 y3 q 3
0
q 0

8 y + y + 8 y y q
1
1
0
1
2
3
=
2 L y0 8 y1 y2 + 8 y3 q 2

0
q 3

yi = q =

q 0
0.0000
q
0.0052
radians = 1

q 2
0.0041

0.0000
q 3
The solution at the end is q0=0.00000 radians, which is exact, while
the value in the center is also exact. The relationship for moment is as
follows:
M yi + 2 + 16 yi +1 30 yi + 16 yi 1 yi 2
=
EI
12h 2
0 y0 M 0
30 32 2
16 31 16
1 y1 M 1
EI
=

12h 2 1 16 31 16 y2 M 2

2 32 30 y3 M 3
0
yi =

+ 16 yi +1 30 and
yi + 16
yi 1 yi 2
M yi + 2Integration
Numerical
Differentiation
131

yi =

=
EI
12h 2
0 y0 M 0
30 32 2
16 31 16
1 y1 M 1
EI

=
12h 2 1 16 31 16 y2 M 2


2 32 30 y3 M 3
0
0 y0 M 0
30 32 2
16 31 16
1 y1 M 1
3EI
=
2
L 1 16 31 16 y2 M 2

2 32 30 y3 M 3
0
30 y0 + 32 y1 2 y2 M 0

3EI 16 y0 31y1 + 16 y2 y3 M 1
=
2
L y0 + 16 y1 31y2 + 16 y3 M 2

2 y1 + 32 y2 30 y3 M 3
M 0
3920
M
676

kip-inches = 1
M 2
1510

2252
M3

The exact solution at the end is M0=4500 k-in, which is a 12.89% error,
while the value in the center is 2250 k-in, which is a 0.11% error. The large
error at the end is due to the fact that the moment drops there, which creates a discontinuity in the equation. Using more segments would reduce
the error, but it would still be more inaccurate than the other values. The
relationship for shear is as follows:
y + 8 yi + 2 13 yi +1 + 13 yi 1 8 yi 2 + yi 3
V
= i+3
EI
8h3
0
0 y0 V0
0 0
13 8 13 8 y V
EI
1 = 1
3
8h 8 13 8 13 y2 V2

0
0 y3 V3
0 0
yi =

0
0 y0 V0
0 0
13 8 13 8 y V
27 EI
1 = 1
3
L 8 13 8 13 y2 V2

0
0 y3 V3
0 0
0
V0

13 y 8 y 13 y + 8 y V
27 EI
1
0
1
2
3
=
3
L 8 y0 + 13 y1 + 8 y2 13 y3 V2

0
V3

8h 8 13 8 13 y2 V2

0
0 y3 V3
0 0
0
0 y0 V0
0 0
13 8 13 8 y V
27 EI
1 = 1
3
8 13 8 13 y2 V2
L
132NUMERICAL STRUCTURAL ANALYSIS

0
0 y3 V3
0 0
0

V0
13 y 8 y 13 y + 8 y V
27 EI
1
0
1
2
3
=
3
L 8 y0 + 13 y1 + 8 y2 13 y3 V2

V3
V0
0
V
07
V0
54.76 i = V1
2
0 kips = V21
V2
25.60
0
V3
0
V3
The exact solution at the end is V0=75.00 k, which is a huge error, while
the value in the center is exact. The large error at the end is due to the fact
that the shear and moment drop dramatically at that point, which creates a
discontinuity in the equation.
The final example of using difference operators to solve differential equations is the critical buckling load of a column. The critical buckling load
of a pinned end column is sometimes included in strength of materials, but
will be derived here. The derivations start with the differential equation of
the elastic curve similar to a beam. The deflected column under the critical
load is shown in Figure3.8.
d2y
M
= y =
EI
dx 2
At any point x along the column, there is a deflection y that will produce an
eccentric moment in the column equal to Pcry. This is used in the equation
of the elastic curve as follows:
Z

Pcr

Pcr
B

A
X
L

Figure3.8. Numeric modeling with difference operators.

Numerical Integration and Differentiation 133

P
d2y
= cr y
2
EI
dx
Pcr
y +
y=0
EI
This is a second-order, linear, ordinary differential equation with the following general solution.
y = Acos ( kx ) + Bsin ( kx )

where k =

Pcr
EI

The constants A and B can be evaluated using the boundary condition. At


x=0, y=0, the equation becomes:
0 = Acos ( k 0) + Bsin ( k 0)
This equation yields A=0 and at x=L, y=0, the equation becomes:
0 = Bsin ( kL)
This condition is true when Bsin(kL)=0, which can only be true for three
conditions as follows:
B=0
kL=0
kL=p, 2p, 3p, ,=np

No deflection
No load
where n=1, 2, 3,

Therefore, the following can be found and is the critical buckling load:
kL =

Pcr
L = np
EI

Pcr 2
L = n 2p 2
EI
n 2p 2 EI
Pcr =
L2

134NUMERICAL STRUCTURAL ANALYSIS

The lowest buckling mode corresponds to a value of n=1, which is the


case of single curvature shown. The other (higher) modes can also be
found with the other values of n.
Pcr =

p 2 EI
L2

The critical bucking load is sometimes written as a critical buckling stress


as follows:
Fcr =

Pcr p 2 EI p 2 EIr 2
p 2E
=
=
=
A
AL2
IL2
L2 2
r

( )

This is known as the Euler buckling stress and was derived by Leonhard
Euler in 1757.
Example 3.12 Column buckling with difference operator
Calculate the critical buckling load, Pcr, for a 25 foot long fixed end column using central difference operator of order of error h2 at 1/6th points.
The column has E=29,000 ksi and I=1000 in4.
This problem has the same model as Example 3.11 except an axial
load is applied instead of a uniform lateral load and is shown in Figure3.9.
The central difference expressions with error of order h2 will be used
to solve for the values. Since the load is known, we will use the fourth
derivative relationship between load and deflection. This can be found
Z

P
A

B
25-0

Y3

Y2

Y1

Y0

Y1

Y2

Y3

Y2

Y1

Y0

Y1

Figure3.9. Example 3.12 Column buckling with difference operator.

Y2

Y3

Numerical Integration and Differentiation 135

by taking two derivatives from previously used deflection and moment


differential equations.
y +

Pcr
y=0
EI

y +

Pcr
y = 0
EI

y 2 yi + yi 1
yi + 2 4 yi +1 + 6 yi 4 yi 1 + yi 2
+ Q i +1
h2
h4
Q
1
0 = 4 ( yi + 2 4 yi +1 + 6 yi 4 yi 1 + yi 2 ) + 2 ( yi +1 2 yi + yi 1 )
h
h
y + Qy = 0 =

The solution process is similar to Example 3.10 and 3.11 except that the
two operators are placed on the model at the location of the unknown
deflections. The value at point 0 of y0 is zero and can be eliminated from
the solutions. This becomes a homogeneous linear algebraic solution set.

7 4 1
1296 4 7 4 + 36 Q

L4
L2

2 8 6

2 1 0 y1 0
1 2 1 y = 0

2
0 2 2 y3 0

144
36
252
L2 2Q L2 + Q

L2

y1 0
144 + Q 252 2Q 144 + Q y = 0
L2
2
L2
L2

y3 0
288
216
72

Q
2
Q

L2
L2
L2
A non-trivial solution to a homogeneous linear algebraic set exists if the
determinant of the coefficient matrix is zero. Therefore, we can find Q by
setting the determinant of the coefficient matrix equal to zero. This will be
done using the basket weave method for a 33 matrix.
252
252
72
216
144
144
0 = 2 2Q 2 2Q 2 2Q + 2 + Q 2 + Q 2
L
L
L
L
L
L
36 252
36 144
72
288
+ 2 2 + Q 2 + 2Q 2 2 2Q 2
L L
L L
L
L
144
144
216

2 + Q 2 + Q 2 2Q
L
L
L

252
144
288
2 2Q 2 + Q 2 + 2Q

L
L
L
252

54432

36Q

41472

576Q

252
252
216
144
144
72
0 = 2 2Q 2 2Q 2 2Q + 2 + Q 2 + Q 2
L
L
L
L
L
L
36 252
36 144 STRUCTURAL
288 ANALYSIS
136NUMERICAL

72
+ 2 2 + Q 2 + 2Q 2 2 2Q 2
L L
L
L L
L
144
144
216

2 + Q 2 + Q 2 2Q
L
L
L

252
144
288
2 2Q 2 + Q 2 + 2Q

L
L
L

36QQ
41472
Q Q 2 2
252
54432 936
41472576576
252
54432
4Q42Q2 4 4 2 2 2Q
0 = 22 22QQ 4 4 2 2 + +
2Q
L

LL
LL
L L
L L
Multiplying the values for like terms yields the following:
504Q
10386 72Q
144

0 = 2 Q 31104 + 2 2Q 2 +
2
L

L
L4
L
18144 144Q
36 41472 576Q
2 + 2Q 2
2
+ 2
L L4
L
L4
L
Multiplying all the values and combining like terms:
0=

1119744 49248
576

Q + 2 Q 2 2Q 3
L6
L4
L

The solution to these cubic equations yields the following general value
of Q:
Q=

Pcr
= 0.0004
EI

Substituting in the particular values for the column of E and Igives


Pcr=11,600 kips. The exact values using the Euler buckling equation with an
effective length factor of k=0.5 for a fixed end column is Pcr=12,721 kips.
This central difference solution has an error of 8.8%.

3.10PARTIAL DIFFERENTIAL EQUATION


DIFFERENCE OPERATORS
More advanced structural analysis problems may require the solution of
partial differential equations. One example is the bending of a plate under
uniform lateral load. Theory of Plates and Shells by Timoshenko and
Woinowsky-Krieger (1959) contains an exact solution to general plates.
The differential relationship for plate bending uses partial differential
equations. Their difference operators can be derived from the basic difference operators using two basic principles. If you add two differential

Numerical Integration and Differentiation 137

equations you simply add the location of the operator at i, known as the
pivot point. To take the product of two differential equations, you must
multiply the value of the operator at iof one differential to each of the
terms in the other differential. Partial differentials also require both
the x and y direction, so they will be written horizontally and vertically.
The following are six examples of creating partial difference operators for
the equation used to solve plate bending. The partial differential equation
is listed first in each of the six examples followed by the basic operators
that represent the equation.
Example 3.13 Partial difference operator

d 4z
d 4z
d 4z 1
+
2
+
=
(1 4 6 4 1)
dx 4
dx 2 dy 2 dy 4 h 4

1
1 1
1
+ 2 * 2 (1 2 1) * 2 2 + 4
h
h h

1
1
= 4 (1 4 6 4 1) + 2 * 4
h

1
2
1


2 4 2 + 1

h4
1 2 1

1
4

6
4

1

1
4

6
4

1

2 8 2

1
= 4 1 8 20 8 1
h

2 8 2

1
1

2 8 2

q d 4z
d 4z
d 4z 1
= 4 + 2 2 2 + 4 = 4 1 8 20 8 1 (3.12)
D dx
dx dy
dy
h

2 8 2

138NUMERICAL STRUCTURAL ANALYSIS

The value D is the flexural rigidity of the plate and is equal to the following:
D=

Et 3

12 1 n 2

The value of Poissons ratio, u, relates the elastic modulus to the shear
modulus and is given as the following from strength of materials:
G=

E
2 (1 + n )

Example 3.14 Partial difference operator


n
n

d 2z
d 2z 1
1
1

=
+
=

n
n
1
2
1
2
1
2

1
+
n
1
(
) 2 2
( )
dx 2
dy 2 h 2
h
h
n

d 2 z
d 2z 1
M x = D 2 + n 2 = 2 1 2 (1 + n ) 1 (3.13)

dy h
dx
n

Example 3.15 Partial difference operator

d 2z d 2z 1
n 2 + 2 = 2 (n
dx
dy
h

2n

1
n ) + 2
h

2 = 1 n
h2
1

d 2 z d 2 z 1
M y = D 2 + 2 = 2 n
dy h
dx

2 (1 + n ) n

2 (1 + n ) n

Example 3.16 Partial difference operator


1
1 0 1
d 2z 1
1
1
=
( 1 0 1) * 0 = 2 0 0 0
dxdy 2h
4h
2h
1 0 1
1

(3.14)

Numerical Integration and Differentiation 139

M xy

1 0 1
d 2z
1

= D (1 n )
= 2 0 0 0
dxdy 4h 1 0 1

M yx

1 0 1
d 2z
1

= D (1 n )
= 2 0 0 0 (3.16)
dxdy
4
h

1 0 1

(3.15)

Example 3.17 Partial difference operator


1
d 3z d 3z
1
1
1
+
=
( 1 2 0 2 1) + ( 1 0 1) * 2 2
dx3 dxdy 2 2h3
2h
h
1

1 0 1
1
1

= 3 ( 1 2 0 2 1) + 3 2 0 2

2h
2h
1 0 1

1 0 1

1
= 3 1 4 0 4 1

2h
1 0 1

1 0 1

d 3 z
d 3z
1
(3.17)
=
Qx = D 3 +

1
4
0

4
1
2
3

2
dx
dxdy
h

1 0 1

Example 3.18 Partial difference operator

2
1
3
3
d z d z 1
1
1

+ 3 = 2 (1 2 1) * 0 + 3 0
2
dx dy dy
2h 2h
h
1
2
1

140NUMERICAL STRUCTURAL ANALYSIS

1 4 1

2
1 2 1

1
1
1

2h 3 0 0 0 + 2h 3 0 = 2h 3 0 0 0
2
1 4 1

1 2 1

1 4 1

d 3z
d 3z
1
Qy = D 2 + 3 = 3 0 0 0 (3.18)
dx dy dy 2h 1 4 1

3.11NUMERIC MODELING WITH PARTIAL


DIFFERENCE OPERATORS
The differential equation and the difference operators from Section 3.10
are used in the following example to solve for plate bending. The forces
on a middle surface plate are shown in Figure3.10.
Example 3.19 Plate bending
Calculate the shear, moment, rotation, and deflection for a 1 inch thick,
20 inch square, fixed end steel plate with a uniformly distributed load
of 100psi using central difference operator of order of error h2 at 1/10th
points. The plate has E=29,000 ksi and u=0.25.

dy
My
My + y dy
Myx
dy
y
Qy
Qy + y dy

Myx +

dx

Mx + Mx
x dx
Mxy
dx
x
Qx
Qx +
dx
x

Mxy +

Figure3.10. Numeric modeling with partial difference operators.

Numerical Integration and Differentiation 141

The plate will be divided into 10 segments in the horizontal and vertical
directions. Since the plate is square with a uniform load and fixed on all
edges, it has horizontal and vertical symmetry. It is also symmetric about
both diagonals of the plate. The symmetry will result in only 21 unique points
on the plate for the 121 possible locations. This is shown in Figure3.11.
The difference operator for Equation 3.12 is repeated here and can be
shown in graphical form in Figure3.12.
1

2 8 2

q d 4z
d 4z
d 4z 1
= 4 + 2 2 2 + 4 = 4 1 8 20 8 1
D dx
dx dy
dy
h

2 8 2

1
The operator is laid upon each of the 21 unique points on the plate resulting in 21 equations involving the deflections. Figure3.13 shows the operator laid upon point 1.
The result will be a non-homogenous set of 21 equations involving
the unknown deflections. Since the plate is fixed at the boundaries, the

13

13

16

17

18

17 16

14

10

10 14

17

19

20

19 17

15

11

11 15

18

20

21

20 18

14

10

10 14

17

19

20

19 17

13

13

16

17

18

17 16

12

12

13

14

15

14 13

10

11

10 9

10

11 10

12

12

13

14

15 14

13

Figure3.11. Example 3.19 Plate bending.

142NUMERICAL STRUCTURAL ANALYSIS


1

-8

-8

20

-8

-8

Figure3.12. Example 3.19 Plate bending.

deflections at points 1 through 6 are zero and may be eliminated. The


matrix equation is shown in Table3.17.
The solution can be found using one of the methods from Chapter2
on non-homogeneous linear algebraic equations. The deflections are
shown in Table3.18.
All of the other values can be found once the deflections are known.
Table3.19 shows the matrix equation for finding Mx and Table3.20 shows
the solution to Mx and the normal stress.

d 2 z
d 2z 1
M x = D 2 + n 2 = 2 1 2 (1 + n ) 1

dy h
dx
n

15

11

11

15

18

20

21

14

10

10

14

17

19

20

13

13

16

17

18

12

31

12

13

14

15

72

28

72

10

11

31

28

10

28

31

72

28

72

10

11

31

12

13

14

15

12

Figure3.13. Example 3.19 Plate bending.

Numerical Integration and Differentiation 143

Table3.17. Example 3.19 Plate bending


22 -16 2 0 0 2 0 0 0 0 0 0 0 0 0
-8 23 -8 1 0 -8 3 0 0 0 0 0 0 0 0
1 -8 21 -8 1 2 -8 2 0 1 0 0 0 0 0
0 1 -8 22 -8 0 2 -8 2 0 1 0 0 0 0
0 0 2 -16 21 0 0 4 -8 0 0 1 0 0 0
2 -16 4 0 0 20 -16 2 0 2 0 0 0 0 0
0 3 -8 2 0 -8 22 -8 1 -8 3 0 0 0 0
0 0 2 -8 2 1 -8 21 -8 2 -8 2 1 0 0
0 0 0 4 -8 0 2 -16 20 0 4 -8 0 1 0
0 0 2 0 0 2 -16 4 0 20 -16 2 2 0 0
0 0 0 1 0 0 3 -8 2 -8 23 -8 -8 3 0
0 0 0 0 1 0 0 4 -8 2 -16 20 4 -8 1
0 0 0 0 0 0 0 2 0 2 -16 4 22 -16 2
0 0 0 0 0 0 0 0 1 0 6 -8 -16 25 -8
0 0 0 0 0 0 0 0 0 0 0 4 8 -32 20

Table3.18. Example 3.19 Plate bending


z7
z8

0.000291121
0.000749271

z9

0.001156404

z10

0.001421248

z11

0.001511939

z12

0.001945784

z13

0.003037557

z14

0.003762980

z15

0.004014203

z16

0.004780224

z17

0.005951021

z18

0.006358843

z19

0.007429735

z20

0.007946380

z21

0.008501580

inches

z7 = q7
h4
z8
q8 * D
z9
q9
z10
q10
z11
q11
z12
q12
z13
q13
z14
q14
z15
q15
z16
q16
z17
q17
z18
q18
z19
q19
z20
q20
z21
q21

2.5

-2.5

1.0

0.0

0.0

0.0

1.3

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

-2.5

1.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.3

0.0

0.0

0.0

1.0

-2.5

1.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.3

0.0

0.0

0.0

1.0

-2.5

1.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.3

0.0

0.0

0.0

2.0

-2.5

1.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.3

0.0

0.0

0.0

0.0

-2.5

1.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

1.0

-2.5

0.0

0.0

0.0

0.0

0.5

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

1.3

0.0

0.0

1.0

-2.5

1.3

0.0

0.0

0.0

0.5

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.3

0.0

0.0

1.0

-2.5

1.0

0.0

0.0

0.0

0.5

0.0

0.0

0.0

Table3.19. Example 3.19 Plate bending

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.3

0.0

0.0

2.0

-2.5

1.0

0.0

0.0

0.0

0.5

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.3

0.0

0.0

0.0

-2.5

1.0

0.0

0.0

0.0

0.5

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

1.0

-2.5

0.0

0.0

0.0

0.3

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

1.3

0.0

1.0

-2.5

1.3

0.0

0.0

0.3

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.3

0.0

2.0

-2.5

1.0

0.0

0.0

0.3

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.3

0.0

0.0

-2.5

1.0

0.0

0.0

0.3

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

1.0

-2.5

0.0

0.0

0.3

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

1.3

2.0

-2.5

1.3

0.0

0.3

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.3

0.0

-2.5

1.0

0.0

0.3

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

2.0

-2.5

0.0

0.3

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

2.5

-2.5

1.3

0.3

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

-2.5

0.3

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

0.0

z21

z20

z19

z18

z17

z16

z15

z14

z13

z12

z11

z10

z9

z8

z7

z6

z5

z4

z3

z2

z1

144NUMERICAL STRUCTURAL ANALYSIS

Numerical Integration and Differentiation 145

Table3.20. Example 3.19 Plate bending


Mx =

0.000000000 k-in s =
x

0.00000

-0.093805542

-0.56283

-0.241431854

-1.44859

-0.372619154

-2.23571

-0.457957773

-2.74775

-0.487180224

-2.92308

-0.134551926

-0.80731

-0.039177480

-0.23506

-0.025067632

-0.15041

-0.036067778

-0.21641

-0.042662612

-0.25598

0.084373209

0.50624

0.258403727

1.55042

0.330357218

1.98214

0.349192954

2.09516

0.460673396

2.76404

0.605975261

3.63585

0.647615145

3.88569

0.774999710

4.65000

0.832219502

4.99332

0.894489044

5.36693

ksi

REFERENCES
Gauss, C.F. 1801. Disquisitiones Arithmeticae. Lipsia, Germany.
Romberg, W. 1955. Vereinfachte numerische Integration. Trondheim, Norway:
Det Kongelige Norske Videnskabers Selskab Forhandlinger.
Salvadori, M., and M. Baron. 1961. Numerical Methods in Engineering. 2nd ed.
Englewood Cliffs, New Jersey: Prentice-Hall.
Simpson, T. 1750. Doctrine and Application of Fluxions. London, England.
Timoshenko, S., and S. Woinowsky-Krieger. 1959. Theory of Plates and Shells.
2nd ed. New York, New York: McGraw-Hill.

CHAPTER 4

Matrix Structural
Stiffness
Building structures are made up of columns, beams, girders, joists, slabs,
walls, shells, and many other components that act together to resist the
loads placed on them. These members can be at various orientations, but
they must be represented in a common mathematical form. The structural
stiffness method is used to represent members, loads, support constraints,
and other components of a building structure in a consistent manner. This
chapter will focus on linearly elastic members subjected to axial, bending,
shear, and torsional forces.

4.1MATRIX TRANSFORMATIONS AND


COORDINATE SYSTEMS
When using the structural stiffness method, the global Cartesian righthand coordinate system will be used to organize the system. This system was developed in 1637 by Ren Descartes (Descartes 1637). We will
denote these three orthogonal axes as X, Y, and Z. You could represent
these as your thumb, fore finger, and middle finger on your right hand.
Individual members may not be at the same orientation as the global system. All members have their own local system represented by x, y, and z.
This coordinate system is also a Cartesian right-hand system with the
xaxis running along the member length. Both of these systems are shown
in Figure4.1.
This chapter provides techniques to reference causes and effects by
coordinate systems and how to manipulate these coordinate systems.
These manipulations allow changes from one system to another and are
called transformations. The two basic transformations that are done in

148NUMERICAL STRUCTURAL ANALYSIS


Z

Figure4.1. Coordinate systems.

the following sections are transmission and rotations. Transmission is


moving effects from one point to another. Rotations are used to re-orient
axes about a specific point. These techniques will be used in determining
the stiffness of a member (local) and developing the global (joint) stiffness solution.

4.2 ROTATION MATRIX


Rotation can take place about any of the three global axes. The following
three examples derive the rotation transformations. The alpha, a, rotation
is a rotation about the global Z axis from the global system to the local
system. Rotation about the Y axis is a beta, b, rotation and rotation about
the X axis is a gamma, g, rotation.
Example4.1 Rotation matrix, a
Derive the alpha, a, rotation matrix.
The following variables are represented in Figure4.2 and are used to
develop a. The location of z remains unchanged since rotation is occurring
about that axis.
a=rotation about global Z axis from the global to the local system
(x0, y0)=global coordinate location
(xl, yl)=local coordinate location

Matrix Structural Stiffness 149


Y

y0

xl

yl

x0

Figure4.2. Example4.1 Rotation, a.

a = x0 cosa
b = y0 sina
c = y0 cosa
d = x0 sina
xl = a + b = x0 cosa + y0 sina
yl = c d = y0 cosa x0 sina
zl = z 0
The equations for x, y, and z can be represented in matrix form as follows:
cosa
sina

sina
cosa
0

0 xo xl
0 yo = yl
1 zo zl

(4.1)

Example4.2 Rotation matrix, b


Derive the beta, b, rotation matrix.
The following variables are represented in Figure4.3 and are used to
develop b. The location of y remains unchanged since rotation is occurring
about that axis.
b=rotation about global Y axis from the global to the local system
(x0, z0)=global coordinate location
(xl, zl)=local coordinate location
a = x0 cos b
b = z0 sin b
c = z0 cos b
d = x0 sin b
xl = a b = x0 cos b z0 sin b
zl = c + d = z0 cos b + x0 sin b
yl = y0

150NUMERICAL STRUCTURAL ANALYSIS


Z

d
zl

c
z0

x0

xl

a = x0 cos
x b
b = z0 sin b

c = z cos b

0
Figure4.3. Example4.2 Rotation,
b.

d = x0 sin b

xl = a b = x0 cos b z0 sin b
zl = c + d = z0 cos b + x0 sin b
yl = y0
The equations for x, y, and z can be represented in matrix form as follows:
cos 0 sin xo xl
0
1
0 yo = yl

sin 0 cos zo zl

(4.2)

Example4.3 Rotation matrix, g


Derive the gamma, g, rotation matrix.
The following variables are represented in Figure4.4 and are used to
develop g. The location of x remains unchanged since rotation is occurring
about that axis.
g=rotation about global X axis from the global to the local system
(y0, z0)=global coordinate location
(yl, zl)=local coordinate location
a = y0 cosg
b = z0 sing
c = z0 cosg
d = y0 sing
yl = a + b = y0 cosg + z0 sing
zl = c d = z0 cosg y0 sing
xl = x0

Matrix Structural Stiffness 151


Z

z0

yl

zl

a = y0 cosg
b = z0 sing

y0

= z0 cosgg.
Figure4.4. Example4.3cRotation,
d = y0 sing

yl = a + b = y0 cosg + z0 sing
zl = c d = z0 cosg y0 sing
xl = x0
The equations for x, y, and z can be represented in matrix form as follows:
0 xo xl
sing yo = yl
cosg zo zl

0
1
0 cosg

0 sing

(4.3)

The rotation matrix from the global to the local system, [R0l], is found
from these rotations about the global Z, Y, and X axes, in that order. The
order of the matrix multiplication goes from right to left. Alpha is first
multiplied by beta and the resultant is multiplied by gamma.

[ R ] = [g ][ b ][a ]
0l

Substituting Equations 4.1 through 4.3 for [g], [b], and [a], respectively,
results in the following:
0
0 cos b 0 sin b cosa sina 0
1

[ R0l ] = 0 cosg sing 0 1 0 sina cosa 0


0 sing cosg sin b 0 cos b 0
0
1
0
0 cosacos b
1
0 cosg sing sina
R
=
[ 0l ]

0 sing cosg cosa sin b

sinacos b
cosa
sina sin b

sin b
0
cos b

cosacos b
sinacos b
sin b

[ R0l ] = cosasin b sing sinacosg sinasin bsing + cosacosg cos b sing


cosa sin b cosg + sina sing sina sin b cosg cosa sing cos b cosg

0 sing cosg sin b

cos b 0

0
0 cosacos b sinacos b
1

cosa
[ R0l ] = 0 cosg sing sina
0 sing STRUCTURAL
cosg cosa sin
b sina sin b
152NUMERICAL
ANALYSIS

sin b
0
cos b

cosacos b
sinacos b
sin b

[ R0l ] = cosasin b sing sinacosg sinasin bsing + cosacosg cos b sing


cosa sin b cosg + sina sing sina sin b cosg cosa sing cos b cosg
Rotation about the axes from the local to the global system is simply a
reverse operation. It can be shown that the resulting rotation matrices for
this transformation are the transpose of the rotations from the global axis
to the local axis. This is known as a symmetric transformation. The following relationships show those basic expressions:
cosa sina 0
[al 0 ] = [a0l ] = sina cosa 0
0
0
1
cos b 0 sin b
T
b
=
b
=
[ l 0 ] [ 0l ] 0 1 0
sin b 0 cos b
T

[g ] = [g ]

l0

0l

0
0
1

= 0 cosg sing
0 sing cosg

The rotation matrix from local to global system, [Rl0], can be found from
the individual rotations or directly from [R0l].

[ Rl 0 ] = [ R0l ]

cosa

[ Rl 0 ] = sina
0

= [a ] [ b ] [g ]
T

sina
cosa
0

cosa cosb
[ Rl 0 ] = sina cosb
sinb

[ Rl 0 ] = [ R0l ]

0 cosb
0 0
1 sinb

0 sinb 1
0
1
0 0 cosg
0 cosb 0 sing

cosa sinb sing sina cosg


sina sinb sing + cosa cosg
cosb sing

cosa cosb

= cosa sinb sing sina cosg


cosa sinb cosg + sina sing

0
sing
cosg

cosa sinb cosg + sina sing


sina sinb cosg cosa sing
cosb cosg

sina cosb
sina sinb sing + cosa cosg
sina sinb cosg cosa sing

sinb
cosb sing
cosb cosg

Matrix Structural Stiffness 153

These rotation transformations will be used later to relate the local member stiffness to the global member stiffness.

4.3TRANSMISSION MATRIX
In addition to knowing the local components of global forces, the effect a
force applied to one end of a member has on the other end is important. It
is often necessary to state the effect at one point in a structural system due
to a cause known to exist at some other point in the system. This is where
the transmission transformation is used. In Figure4.5, the cause of a force
at point 2 is transmitted to the effect at point 1. This is achieved by using
an equivalent static force at 1. In the study of rigid body equilibrium, this
is stated as SF1= SF2.
The six orthogonal forces at a point are shown in Figure4.6. These
are the forces and moments in each direction X, Y, and Z. The moments
are represented with double arrowheads.
The static equivalent force system is found where the moment arm
distances (x2 x1), (y2 y1), and (z2 z1) are measured from the effect point,
1, to the cause point, 2.
Z
Y
2 . (x2,y2,z2)
1 . (x1,y1,z1)

Figure4.5. Transformation locations.

MZ
MY
PZ
1

Figure4.6. Orthogonal forces.

PY
PX

MX

154NUMERICAL STRUCTURAL ANALYSIS

Fx1 = Fx 2 Px1 = Px 2
Fy1 = Fy 2 Py1 = Py 2
Fz1 = Fz 2 Pz1 = Pz 2

M x1 = M x 2 M x1 = M x 2 Py 2 ( z2 z1 ) + Pz 2 ( y2 y1 )

M y1 = M y 2 M y1 = M y 2 + Px 2 ( z2 z1 ) Pz 2 ( x2 x1 )

M z1 = M z 2 M z1 = M z 2 Px 2 ( y2 y1 ) + Py 2 ( x2 x1 )

We can simplify the moment arm distances as x, y, and z. The concept of


transmitting cause to effect may be denoted in matrix form as [T]:

0
0
1
0
1
0

0
0
1

0 z y
z
0 x

0
y x

0
0
0
1
0
0

0 Px 2 Px1
0 Py 2 Py1
0 Pz 2 Pz1
=

0 M x 2 M x1
0 M y 2 M y1

1 M z 2 M z1 

0
0
0
0
1
0

(4.4)

Example4.4Transmission matrix
Determine the transmission matrix for the coplanar XY system from the
origin end, i, to other end, j.
The cause end is i as shown in Figure4.7 and the effect end is j at the
right end a distance L away.
Y

i = cause

j = effect

Figure4.7. Transformation effects.

Matrix Structural Stiffness 155

The distance x is measured from the cause location to the effect


location.

x = ( xc xe ) = xi x j = L
This can be substituted into Equation 4.4 and the unneeded distances and
forces are removed to get a 33 matrix involving only Px, Py, and Mz.
1 0 0 0 0 0 Px 2 Px1
0 1 0 0 0 0 P P

y 2 y1
0 0 1 0 0 0 Pz 2 Pz1
=

0 0 0 1 0 0 M x 2 M x1
0 0 L 0 1 0 M y 2 M y1

0 L 0 0 0 1 M z 2 M z1
1 0 0 Pxi Pxj
0 1 0 P = P
yi yj

0 L 1 M zi M zj
The axial force in the member, Px, is a direct transmission, whereas the
shear, Py, and bending, Mz, forces are linked. The transmission for just the
shear and bending will be used later in the chapter and can be written as
follows:
1 0 Pyi Pyj
L 1 M = M
zi zj

(4.5)

A similar transmission matrix can be derived for shear and bending in the
XZ system.
1 0 Pzi Pzj
L 1 M = M
yi yj

(4.6)

4.4 AREA MOMENT METHOD


Area moment and the next five sections are review topics on structural
analysis. These topics will be used to derive many of the equations and

156NUMERICAL STRUCTURAL ANALYSIS

relationships for stiffness methods of analysis. The area moment method


of analysis is based on the basic equation of beam bending, f = M/EI,
and consists of two theorems. The theorems were developed by Christian
Otto Mohr in 1874 (Timoshenko 1953). The first theorem of area moment
states the change in slope between two points on the elastic curve, 1 and 2,
is equal to the area of the M/EI diagram between the points 1 and 2.
2

q12 =
1

M
dx
EI

The second theorem of area moment states that the tangential deviation
of a point, 1, from a tangent to the elastic curve at point 2 (the tangent)
is equal to the moment of the area of the M/EI diagram between points
1and2, taken about point 1 (the point). Note that the moment of the area
is the area times a distance, x1, from point 1.
2

t12 =
1

M
x1dx
EI

Example4.5 Area moment


Find the reactions on the following propped end, uniformly loaded beam
shown in Figure4.8 using the area moment method.
Z
W
A

Figure4.8. Example4.5 Area moment.

A free-body diagram of the beam is drawn in Figure4.9.


W
MB
RA

Figure4.9. Example4.5 Area moment.

RB

Matrix Structural Stiffness 157

Moment divided by EI, M/EI, diagrams are drawn by parts. Moment


diagrams by parts assume each force is acting about point B as if it was the
only force. The vertical reaction on the left, RA, is used as the redundant
force. The method of superposition will be used to solve for the assumed
redundant, RA. The reactions at the fixed-end on the right can be written in
terms of the load, w, as follows from equilibrium on Figure4.9 (ignoring RA):
wL2
L
M B = 0 = wL M B M B =
2
2
Fy = 0 = wL + RB RB = wL
The M/EI diagram and the deflected shape for the uniformly distributed
load, w, is shown in Figure4.10.
w
wL2
2

tAB
wL
wL2
2EI

3L/4

Figure4.10. Example4.5 Area moment.

The reactions at the fixed-end on the right can be written in terms of the
vertical reaction, RA, as follows from equilibrium on Figure4.9 (ignoring
the distributed load):
M B = 0 = R A L M B M B = R A L
Fy = 0 = RA + RB RB = RA
The M/EI diagram and the deflected shape for the vertical reaction, RA, is
shown in Figure4.11.

tAB

RAL
RA

RA
2L/3

Figure4.11. Example4.5 Area moment.

RA L
EI

158NUMERICAL STRUCTURAL ANALYSIS

Since point A is pinned, we know that the deflection at point A, DA, is


equal to zero. We can superimpose Figures 4.10 and 4.11. Applying the second area moment theorem, the following can be written. It should be noted
that the tangent to the deflected shape is at B and therefore the point is at A:
B

M
x dx = Ax1
EI A
A

A = 0 = t AB =

1 wL2 3L 1 RA L 2 L
3wL
0 = L
+ L
RA =

2 EI
3
8
3 2 EI 4
From statics on the original free-body diagram in Figure4.9, the other
reactions at B can be found.
Fy = 0 = RA + RB wL RB =
M B = 0 = R A L M B +

5wL
8

wL2
wL2
MB =
2
8

4.5CONJUGATE BEAM METHOD


The conjugate beam method is based on the equation of beam bending
f= M/EI. The method was developed by Heinrich Mller-Breslau in 1865
(Mller-Breslau 1875). A conjugate beam can be summarized as an imaginary beam equal in length to the real beam. In the imaginary beam, the
shear at the conjugate support is equal to the slope of the real support.
Also, the moment at the conjugate support is equal to that of the deflection
at the real support. The conjugate beam is loaded with the M/EI diagram
from the real beam. A summary of the more common support conditions
is shown in Figure4.12.
Take the pinned condition for example. In the real beam, the rotation is
unknown (exist) and deflection is equal to zero. Therefore, in the conjugate
beam, the shear is unknown (exist) and the moment is equal, thus creating
a pinned connection. This condition does not change from the real beam to
the conjugate beam. However, this is not true for all cases. For example,
if the real beam is fixed, the rotation is zero and the deflection is zero. The
result is zero moment and shear in the conjugate that yields a free end.
Example4.6 Conjugate beam
Draw the conjugate beam for each of the real beams in Figure4.13.

Matrix Structural Stiffness 159


REAL

CONJUGATE

Pin

Pin

Roller

Roller

Fixed

Free

Free

Fixed

Internal pin

Hinge

Internal roller

Hinge

Hinge

Internal roller

Figure4.12. Conjugate versus real supports.


REAL

CONJUGATE

Figure4.13. Conjugate versus real beams.

Example4.7 Conjugate beam


Find the slope and deflection at the free end of the cantilever beam in
Figure4.14 by the conjugate beam method.
A free-body diagram of the real beam is drawn with the reactions at
the fixed-end found from static equilibrium. The conjugate beam is drawn

160NUMERICAL STRUCTURAL ANALYSIS

observing that the shear in the conjugate beam equals the rotations in the
real beam and the moment in the conjugate equals the deflection in the
real beam. Therefore, the fixed-end of the real beam becomes free in the
conjugate beam and the free end in the real beam becomes fixed in the
conjugate beam. The conjugate beam is loaded with the M/EI diagram of
the real beam. This is a triangular moment diagram due to the applied load
P divided by EI. The last beam in Figure4.14 is a free-body diagram of the
conjugate beam with the area of the M/EI diagram equated to a point load
located at the centroid of the diagram. Static equilibrium can then be performed on the conjugate beam to find shear and moment at point B, which
represents rotation and deflection in the real beam shown in Figure4.15.
P
A

L/2

L/2

P
P

M=PL/2

L/2

L/2

L/2

L/2

PL
2EI
L/6

5L/6

MB

VB

PL2
8EI

Figure4.14. Example4.6 Conjugate beam.

Fy = 0 =

qB =

PL2
PL2
VB VB =
8 EI
8 EI

PL2
8 EI

PL2 5 L
5 PL3
+ MB MB =
8 EI 6
48 EI
3
5 PL
B =
48 EI
M B = 0 =

Matrix Structural Stiffness 161

B
B

Figure4.15. Example4.6 Conjugate beam.

4.6 VIRTUAL WORK


Similar to the conjugate beam method, the virtual work method can be
used to obtain the displacement and slope at a specific point on a structure. The method was formulated by Gottfried Leibniz in 1695 (Leibniz
1695). Virtual work is based on internal strain energy from the member
and external work done by the forces. There are two ways to apply the
virtual work method. The first is to apply a virtual (unit) displacement
to find a real force. The second is to apply a virtual (unit) force to find a
real displacement. The force and the displacement in either case are in the
same direction. The basic equation to find a real displacement based on a
virtual force is as follows:
L

q or =
0

mM
dx
EI

(4.7)

The value of M is the moment equation due to the real loads on the structure. The value of m is the moment equation of the virtual load. Rotation or
deflection may be found depending on whether a virtual moment or force
is applied at the point under consideration.
Example4.8 Virtual work
Determine the vertical deflection at the free end of the uniformly loaded,
cantilever beam in Figure4.16 using the virtual work method.
A free-body diagram is drawn of the right-hand side of the beam to
determine the internal moment in the beam. The uniformly distributed
load is represented as a point load equal to the area under the load and
located at the centroid of the area.
From static equilibrium, we can determine the internal moment at any
point x measured from the right end of the beam.
M=

wx 2
2

162NUMERICAL STRUCTURAL ANALYSIS


w

L
wx
x/2
M= - wx
2

Figure4.16. Example4.8 Virtual work.

A virtual force is applied at the point and in the direction that the deflection is desired. This is shown in Figure4.17 along with a free-body diagram of the right-hand side of the beam. In this case, the internal moment
is represented as m.
w

1 kN

L
1 kN

m= -1x

Figure4.17. Example4.8 Virtual work.

From static equilibrium, we can determine the internal moment at any


point x measured from the right end of the beam.
m = 1x
The values of M and m can be substituted into Equation 4.7 to determine
the deflection at point B at the free end of the cantilever.

Matrix Structural Stiffness 163

wx 2
L
2
wx3
wL4
dx =
dx =
2 EI
8 EI
EI
0

1x )
L (

mM
dx =
EI
0
0

B =

4.6.1 VISUAL INTEGRATION


The virtual work method can also be done using the moment diagrams of
the real and virtual loads. This is known as visual integration. The moment
diagrams for the real (Q) and virtual (q) loads are constructed. The integral
is equal to the area of the M/EI diagram multiplied by the value of the m
diagram taken at the centroid of the M/EI diagram.
Example4.9 Visual Integration
Determine the deflection at C in inches for the simply supported beam
shown in Figure4.18 using the virtual work method and visual integration. Use E=2000 ksi and I=2000 in4.
8k
40 k-ft
A

C
8ft

B
16ft

Figure4.18. Example4.9 Visual integration.

Solve for the reactions due to the applied loads. The free-body diagram is shown in Figure4.19.
M B = 0 = 8k (16 ft ) Ay ( 24 ft ) + 40k ft Ay = 7 k
Fy = 0 = Ay + By 8k By = 1k
The shear and moment diagrams are constructed in Figure4.19. The
moment diagram is broken into unique areas that will be continuous over
the area of the virtual moment diagrams shown in Figure4.20.
Apply a virtual force dQ = 1 at point C as shown in Figure4.20. Solve
for the reactions similar to the real loads. Draw the shear and moment
diagram locating the centroid of the real moment diagrams on the virtual
moment diagram.

164NUMERICAL STRUCTURAL ANALYSIS


8k
40 k-ft
C
7k

8ft

1k

16ft

7
V(k)

2(8ft)/3

-1
2(16ft)/3

56

16ft/2
40

M(k-ft)

Figure4.19. Example4.9 Visual integration.

Q=1

C
2
3

8ft

3.556

1
3

16ft
16
3

3.556
2.667

m(k-ft)

Figure4.20. Example4.9 Visual integration.

M B = 0 = 1k (16 ft ) Ay ( 24 ft ) Ay =
1
Fy = 0 = Ay + By 1k By = k
3

2
k
3

The values of M and m can be substituted into Equation 4.7 to determine


the deflection at point C. The integral is the product of the area M/EI diagram and the value of the m diagram at the centroid of the M/EI diagram.

Matrix Structural Stiffness 165


L

mM
dx
EI
0

dQ C =

1
1 (8 ft ) 56k -ft (3.556k -ft ) + (16 ft )16k -ft (3.556k -ft )
2
2

EI

+ (16 ft ) 40k -ft ( 2.667k -ft )


2
3
2958k -ft
1k C =
EI
2958 1728in3 /ft 3
2958k -ft 3
C =
=
= 1.278in
EI
2000k /in 2 2000in 4

1k C =

4.7CASTIGLIANOS THEOREMS
In 1879 Alberto Castigliano published his two theorems on elastic structures that are known as Castiglianos Theorems (Castigliano 1879). The
first theorem states that the first partial derivative of strain energy with
respect to a particular deflection component is equal to the force applied at
the point and in the direction corresponding to that deflection component.
This may be written in mathematical terms as shown in Equation 4.8. The
second theorem is used more often in statically indeterminate structural
analysis and states that the first partial derivative of strain energy with
respect to a particular force is equal to the displacement of the point of
application of that force in the direction of its line of action. This is shown
in Equation 4.9. The equations are written in terms for flexural energy,
M/EI, of a particular rotation, qA, and moment, MA, relationship and in
terms of a particular deflection, DA, and force, PA, relationship. They can
be written for any elastic force and deformation relationship.
L

M A = M
0

q A = M
0

M dx
M dx
and PA = M
qA EI

A EI
0
L

M dx
M dx
and A = M
M A EI
PA EI
0

(4.8)

(4.9)

Example4.10 Castiglianos second theorem


Determine the deflection of the beam in Figure 4.21 at point B using
Castiglianos second theorem.
Since the deflection at point B is desired, a force P will be placed at B.
This will be the particular force in the direction of the desired deflection.

166NUMERICAL STRUCTURAL ANALYSIS


w

L
P

Figure4.21. Example4.10 Castiglianos second theorem.

A free-body diagram of the right-hand portion of the beam is shown in


Figure4.22. The moment equation is found from equilibrium on the freebody. Also, the partial derivative of the moment equation is taken with
respect to the particular force P. These are substituted into Equation 4.9
to find the deflection at point B.
wx

x/2
M
V

Figure4.22. Example4.10 Castiglianos second theorem.

M B = 0 = M wx
M
= x
P
L

B = M
0

B =

x
wx 2
Px M =
Px
2
2

L
L
wx 2
wx 3

dx
M dx
dx
=
Px ( x )
=
+ Px 2
P EI 0 2
EI 0 2

EI

wL4 PL3
+
8 EI 3EI

Since P is a fictitious applied load placed on the beam only to find the
deflection, it can be removed and the deflection at B becomes the following:
B =

wL4
8 EI

Matrix Structural Stiffness 167

Example4.11 Castiglianos second theorem


Find the reactions for the propped end beam loaded as shown in Figure4.23.
Since this is a statically indeterminate beam, a redundant force will
be applied at point A. The force will be represented as P but is actually the
real vertical reaction on the left end RA. The last drawing in Figure4.23 is
a free-body diagram of a left-hand section of the beam to find the moment
equation by equilibrium.
w
Z

X
A

B
L
w
2

wL -PL
6
wL -P
2

P
wx
L

M
x

Figure4.23. Example4.11 Castiglianos second theorem.

A free-body diagram for the total beam is drawn in Figure4.23 with


reactions at the fixed end written in terms of the redundant force P. These
reactions are found by equilibrium.
M B = 0 = M B PL +

MB =

1
L
wL
2
3

wL2
PL
6

(4.10)

1
Fy = 0 = RB + P wL
2
RB =

wL
P
2

(4.11)

168NUMERICAL STRUCTURAL ANALYSIS

The partial derivative of the moment equation is taken with respect to


the particular force P. These are substituted into Equation 4.9 to find the
forceP, which is the reaction at point A. Note the deflection at point A is
set to zero in Castiglianos equation.
M = 0 = M Px +
M
= x
P
L

A = 0 = M
0

1 wx x
wx 3
x
M =
Px
2 L 3
6L

L
L
wx 4
wx 3
dx

M dx
dx
=
=
+ Px 2
Px ( x )
P EI 0 6 L
EI 0 6 L
EI

wL5 PL3
0=
+
3
30 L
wL
P = RA =
10
This result can be substituted back into Equations 4.10 and 4.11 to find the
reactions at A.
2 wL
5
wL2
MB =
15
RB =

4.8 SLOPE-DEFLECTION METHOD


The slope-deflection method is a stiffness method that includes flexural or
bending stiffness. It was introduced in 1915 by George A. Maney (Maney
1915). In slope-deflection, moments at the end of a member are expressed
in terms of the rotations at the ends and the fixed-end moments due to the
loads. Once the expressions for the moments at the member ends are written, the joint moments are equated to zero and the unknown moments are
found from the system of equations. The basic slope-deflection equations
are shown in Equations 4.12 and 4.13. These are for the i end and j end of
a member, respectively.
M i = FEM ij +

4 EI
2 EI
6 EI
fi +
fj
b
L
L
L

(4.12)

Matrix Structural Stiffness 169

M j = FEM ji +

2 EI
4 EI
6 EI
fi +
fj
b
L
L
L

(4.13)

In these two equations, f is the rotation at the joint and b is the lateral
translation between the ends divided by the length of the member. It will
be seen later that the values 4EI/L, 2EI/L, and 6EI/L are flexural stiffness
terms. The FEM terms are the fixed-end moments due to the loads on the
member. A special case may be used if one end of the member is pinned.
Equation 4.14 is for the i end of a member when the j end is pinned.
M i = FEM ij +

FEM ji
2

3EI
(fi b )
L

(4.14)

Example4.12 Slope-deflection
Determine the moments at the ends of the members of the continuous
beam in Figure4.24 using the slope-deflection equations.
Z

8 kips
1.2 kips/ft

8 kips

B
20ft

5ft

5ft

Figure4.24. Example4.12 Slope-deflection.

The fixed-end moments due to the loads on each span are computed
and can be found in most elementary structural analysis textbooks.
1.2k /ft ( 20 ft )
wL2
=
= 40k ft
12
12
2

FEM AB =

wL2 1.2k /ft ( 20 ft )


= 40k ft
=
12
12
2

FEM BA =

8 (5 ft ) (5 ft )
Pb 2 a
= 10k ft
=
2
L
(10 ft )2
2

FEM BC =

Pa 2b 8 (5 ft ) (5 ft )
= 2 =
= 10k ft
L
(10 ft )2
2

FEM CB

170NUMERICAL STRUCTURAL ANALYSIS

The slope-deflection equations are applied to each span. In this case,


the normal equations (4.12 and 4.13) are used for span AB and a special
case (4.14) is used for span BC since support C is a roller. It should
be noted that there are no rotations at support A and therefore fA =0.
Also, since there is no translational movement between the ends,
b is zero.
4 EI
2 EI
2 EI
EI
fA +
f B = 40 +
f B = 40 + f B
L
L
20
10
4 EI
4 EI
2 EI
EI
= FEM BA +
fA +
f B = 40 +
f B = 40 + f B
20
5
L
L
FEM CB 3EI
10 3EI
3EI
= FEM BC
+
f B = 100 +
f B = 15 +
fB
2
L
2 10
10

M AB = FEM AB +
M BA
M BC

Equilibrium equations are written at each joint that has a real rotation. In
this case, that is only joint B.
M BA + M BC = 0 = 40 +

fB =

50
fB
EI

EI
3EI
f B 15 +
fB
5
10

Substituting the value of fB back into the moment equation will result in
the final member-end moments.
M AB = 40 +
M BA = 40 +

EI 50
= 45k ft
10 EI

EI 50
= 30k ft
5 EI

M BC = 15 +

3EI 50
= 30k ft
10 EI

4.9 MOMENT-DISTRIBUTION METHOD


The moment-distribution method is an iteration process that uses the
same basic assumptions and equations as the slope-deflection method.
Moment-distribution was developed by Hardy Cross in 1930 (Cross
1930). The difference between the two is that at each joint the fixed-end
moments are first summed and distributed to each member in proportion to

Matrix Structural Stiffness 171

their flexural stiffness (stiffness factor). Then, the member-end moments


are carried over to their far ends by the carryover factor. The process is
repeated and continues until the amount of moment being distributed
becomes significantly small.
Many factors are used with the moment-distribution method. The first
is the member stiffness factor and is the amount of moment required to
rotate the end of a beam 1 radian. This is actually the definition of stiffness, force due to unit motion. The far end of the beam that is rotating is
fixed. We will derive this expression in the next section on elastic member
stiffness.

K=

4 EI
L

The joint stiffness factor is the sum of all the member stiffness factors for
the members connected at a joint.
K T = K

The distribution factor for each member-end at a joint is the member stiffness factor divided by the joint stiffness factor.
DF =

K
K
=
KT K

If a member is connected to a support and not to other members, the distribution factor is dependent on the support type. If the support is fixed
against rotation, then DF=1. If the support is free to rotate, then DF=0.
The member relative stiffness factor is used when a continuous beam
or frame is made from the same material when calculating the distribution
factor. This can be used in place of the member stiffness factor for calculation of the other factors.
KR =

I
L

The final factor is the carry-over factor, which represents the fraction of a
moment that is carried over from one end of a member to the other. If the
member is prismatic, then the ratio of the far end moment to the near end
moment of a member is one-half ().

172NUMERICAL STRUCTURAL ANALYSIS

Finally, if a member is connected to a support, but is free to rotate at


the support, a modified member stiffness factor can be used. In this modified method, the support joint is moment balanced then carried over and
no further calculations are performed at that joint.
K =

3K
4

Example4.13 Moment-distribution method


Determine the moments in the beam shown in Figure4.25 by the
moment-distribution method.
The modulus of elasticity, E, and moment of inertia, I, are constant
for the beam.
Since E and I are constant, the member relative stiffness factor can be
used and we can use a unit value for I.
K AB =

I AB
1
=
= 0.05
LAB 20

K BC =

I BC
1
=
= 0.10
LBC 10

At fixed support A, the distribution factor for member AB is 0 and at the


roller support C, the distribution factor for member BC is 1. For the internal roller, the distribution factor for each member must be calculated.
DFAB =

K AB
0.05
=
= 0.333
K AB + K BC 0.05 + 0.10

DFBC =

K BC
0.1
=
= 0.667
K AB + K BC 0.05 + 0.10

8 kips
1.2 kips/ft

8 kips

B
20ft

Figure4.25. Example4.13 Moment-distribution.

5ft

C
5ft

Matrix Structural Stiffness 173

The fixed-end moments are calculated for each member load.


2

FEM AB

1.2k /ft ( 20 ft )
wL2
=
=
= 40k ft
12
12
2

FEM BA =

wL2 1.2k /ft ( 20 ft )


=
= 40k ft
12
12
2

FEM BC =

8 (5 ft ) (5 ft )
Pb 2 a
=
= 10k ft
2
L
(10 ft )2
2

FEM CB

Pa 2 b 8 (5 ft ) (5 ft )
= 2 =
= 10k ft
L
(10 ft )2

The moment-distribution is shown in Table4.1. In the first line, the fixedend moments are recorded. In the second line, the distribution factor is
multiplied times the negative sum of the fixed-end moments. This is a distribution. In each subsequent one-half, the end moment is carried over the
opposite end of the members. This is a carry-over. Then, the unbalanced
moments at the joint are distributed again and carried over again.
Table4.1. Example4.13 Moment-distribution
0.00

0.333

0.667

-40.00

40.00

-10.00

10.00 Fixed-end

1.00 Distribution

Factor

0.00

-10.00

-20.00

-10.00 Distribution

-5.00

0.00

-5.00

-10.00 Carry-over

0.00

1.67

3.33

10.00 Distribution

0.83

0.00

5.00

1.67 Carry-over

Moment

0.00

-1.67

-3.33

-1.67 Distribution

-0.83

0.00

-0.83

-1.67 Carry-over

0.00

0.28

0.56

1.67 Distribution

0.14

0.00

0.83

0.28 Carry-over

0.00

-0.28

-0.56

-0.28 Distribution

0.14

0.00

-0.14

-0.28 Carry-over

0.00

0.05

0.09

0.28 Distribution

0.02

0.00

0.14

0.05 Carry-over

0.00

-0.05

-0.09

-0.05 Distribution

-0.02

0.00

-0.02

-0.05 Carry-over

-45.00

30.00

-30.02

-0.05 Final Moments

174NUMERICAL STRUCTURAL ANALYSIS

Example4.14 Moment-distribution
Determine the moments in the beam shown in Figure4.25 by the modified
moment-distribution method.
There are two primary differences when using this method. First, use
a modified member stiffness factor for member BC.
K BC
=

3K BC 3 ( 0.1)
=
= 0.075
4
4

The modified factor must be used to recalculate the distribution factors.


DFAB =

K AB
0.05
=
= 0.40
K AB + K BC 0.05 + 0.75

DFBC =

K BC
0.075
=
= 0.60
K AB + K BC 0.05 + 0.075

Second, the fixed-end moment at the pinned support is balanced then carried over to the far end before the moment-distribution process begins.
The process is shown in Table4.2.
Table4.2. Example4.14 Moment-distribution
0.00
0.400 0.600
1.00 Distribution
40.00 -10.00
10.00 Fixed-end
-40.00
-5.00
-10.00 Balance @ C
0.00
Distribution
-10.00 -15.00
0.00
0.00
Carry-over
-5.00
30.00 -30.00
0.00 Final Moments
-45.00

Factor
Moment
1
1

4.10 ELASTIC MEMBER STIFFNESS, X-Z SYSTEM


The stiffness method for analyzing building structures is widely used by
engineers and commercial computer structural analysis programs. The
method was developed 1934 and 1938 by Arthur Collar (Lewis et al.
1939). The basic definition of stiffness is the force due to a unit deformation. Flexibility is the reciprocal or inverse of stiffness and is defined as
the deformation due to a unit force. Either of these principles can be used
to find the behavior of structural members due to motion and loads. In this
section, the elastic member stiffness for a linear element will be derived.

Matrix Structural Stiffness 175


P

Figure4.26. Elastic stiffness.

The fixed-end forces on a member can also be found by similar methods.


When discussing the flexibility and stiffness of a member, an elastic spring
that is axially loaded can be considered. This is equivalent to the axial
properties of a linear element used in structures. Figure4.26 shows an
elastic spring in an un-deformed and deformed position.
The displacement of the spring is directly proportional to the applied
axial load. This is the basic flexibility equation and is written as follows:
= fP
In this equation, f is the flexibility of the spring and P is the applied axial
load in the direction of the length of the spring. This can also be written in
terms of stiffness using a displacement of 1 unit.
P = K
This equation is the general equation for stiffness. P represents all the
known forces, D represents the unknown rotations and deflections, and K
is the stiffness matrix for a member or structure. The entire system of a
structure can be modeled into a set of simultaneous equations written in
the following form and will be expanded in the next section and chapter:

[ K ][ ] = [ P ]
The derivation of elastic member stiffness in the X-Z system will be
derived in the two following examples using the conjugate beam method
and area moment method.
Example4.15 qiy stiffness
Derive the qiy stiffness using the conjugate beam method for a linear
member.
A free-body diagram is shown in Figure4.27 with an imposed rotation of 1 unit on the i-end of the member. The moments are assumed in the

176NUMERICAL STRUCTURAL ANALYSIS


Pjz

Piz

Miy

Mjy

iy
Miy
EIy
PizL
EIy

Figure4.27. Example4.15 qiy stiffness.

positive y direction using the right-hand rule and the Cartesian right-hand
coordinate system. The forces are shown consistent with the deformation.
The moment diagram divided by EI is shown for the reaction forces on the
i-end of the member.
The conjugate beam can be constructed for the two basic assumptions. The shear in the conjugate is equal to the slope of the real beam, and
the moment of the conjugate is equal to the deflection of the real beam.
Viconj = qireal = qiy
M iconj = ireal = 0
Since moment in the conjugate does not exist but shear does, the conjugate
beam is pinned on the i-end.
V jconj = q jreal = 0
M jconj = jreal = 0
Since both the shear and moment in the conjugate do not exist, the conjugate beam is free on the j-end. The resulting conjugate beam is shown in
Figure4.28.

Figure4.28. Example4.15 qiy stiffness.

The conjugate beam method can be applied to find the reactions at the
i-end of the conjugate beam, which are equal to deformations at the i-end
of the real beam. The load from Figure4.27 is applied to the conjugate
beam in Figure4.28.

Matrix Structural Stiffness 177

Viconj = qireal = qij = 0 qiy =

M iy L
EI y

Piz L2
2 EI y

M iy L L Piz L2 2 L
M iconj = ireal = 0 =
+

EI y 2 2 EI y 3
Solving the second equation for Piz in terms of Miy and then substituting
into the first equation, the stiffness value can be found.
Piz =

3M iy

qiy =
M iy =
Piz =

2L
M iy L
EI y

4 EI y

L
6 EI y
L2

3M iy L
4 EI y

M iy L
4 EI y

qiy

(4.15)

qiy

It should be noted that the force Piz was actually shown as negative in the
original free-body diagram so it does have a negative value for stiffness.
Piz =

6 EI y
L2

qiy

(4.16)

Example4.16 Diz stiffness


Derive the Diz stiffness using the area moment method for a linear member.
A free-body diagram is shown in Figure4.29 with an imposed deflection of 1 unit on the i-end of the member. The moments are assumed in the
negative y direction using the right-hand rule and the Cartesian right-hand
coordinate system. The forces are shown consistent with the deformation.
The moment diagram divided by EI is shown for the reaction forces on the
i-end of the member.
Since both ends of the beam are fixed for rotation, the change in rotation from the i-end to the j-end is zero. This is the area under the M/EI
diagram between those points.

qij = 0 = q j qi =
Piz =

2M iy
L

M iy L
EI y

Piz L2
2 EI y

178NUMERICAL STRUCTURAL ANALYSIS


Piz
Pjz
iz

Mjy

Miy
Miy
EIy
PizL
EIy

Figure4.29. Example4.16 iz stiffness.

The tangential deviation of a point at the i-end from a tangent to the curve
on the j-end is the implied deflection. This is equal to the moment of the
area of the M/EI diagram about the point at the i-end.
j

tij = iz =

M iy L L Piz L2 2 L
M
xi dx =
+

EI
EI y 2 2 EI y 3

Substituting the first equation for Piz into the Diz equation results in one
of the stiffness terms. The second term is found by substituting the first
stiffness term back into the Piz equation. Note that Miy was assumed as
negative in the free-body diagram, so the sign must be switched.

iz =

M iy L2
6 EI y
6 EI y

M iy =

Piz =

L2

12 EI y
L3

iz

iz

(4.17)

(4.18)

The four terms given in Equations 4.15 through 4.18 are the flexural stiffness terms for the forces at the i-end due to motions at the i-end. This is
denoted as stiffness matrix [Kii] in Equation 4.19. The stiffness equation
and matrix form of this are as follows:

[ K ][ d ] = [ F ]
ii

6 EI y
12 EI y
2

3
L iz Piz
L

=
4 EI y qiy M iy
6 EI y
2

L
L
6 EI y
12 EI y
2

Matrix Structural Stiffness 179

[ K ][ d ] = [ F ]
ii

6 EI y
12 EI y
2

3
L iz Piz
L

=
4 EI y qiy M iy
6 EI y

L
L2
6 EI y
12 EI y
2

3
[ Kii ] = 6LEI 4 EIL

y
y
2

L
L

(4.19)

The transmission matrix derived in Section 4.3 for the X-Z system can be
used to find the forces at the j-end.
1 0
1

[T ] = L

From statics equilibrium, the forces at the i-end transmitted to the j-end
plus the forces at the j-end are equal to zero. The forces at the j-end due to
motions at the i-end are denoted as stiffness matrix [Kji].

[T ][ Kii ] + K ji = 0
K ji = [T ][ K ii ]
6 EI y
12 EI y
12 EI y
2

3
1 0
L 1 0 L3
L
K ji =
=

4 EI y L 1 6 EI y
L 1 6 EI y
2

2
L
L
L
12 EI y 6 EI y
3

L
L2
K ji =
6 EI y 2EI y
2

L
L

6 EI y

L2
4 EI y

(4.20)

For a prismatic member, it can be shown that the stiffness matrix is symmetric about the main diagonal. This results in stiffness [Kij] being equal
to the transpose of [Kji], where the forces at the j-end due to motions at the
i-end are denoted as stiffness matrix [Kji].

180NUMERICAL STRUCTURAL ANALYSIS


T

12 EI y 6 EI y
12 EI y
3

3
2
T
L
L
L

K ij = K ji =
=
6 EI y 2 EI y
6 EI y
2

L
L

L2
6 EI y
12 EI y
2
3
L
L
K ij =
2 EI y
6 EI y

L
L2

6 EI y

L2
2 EI y

(4.21)

From statics equilibrium, the forces at the i-end transmitted to the j-end
plus the forces at the j-end are equal to zero. The forces at the j-end due to
motions at the j-end are denoted as stiffness matrix [Kjj].

[T ] Kij + K jj = 0
K jj = [T ] K ij
12 EI y
3
1 0
L

K jj =
6 EI
1
L

L2
12 EI y 6 EI y
3

L2
K jj = L
6 EI y 4 EI y
2

L
L

6 EI y
12 EI y
3
2
L
L 1 0
=
2 EI y L 1 6 EI y

L
L2

6 EI y

L2
2 EI y

(4.22)

Adding the axial stiffness terms that are derived in most strength of materials textbooks, the entire coplanar X-Z frame stiffness matrix can be
found.
Ax E
ix
L
AE
Pjx = x ix
L
Ax E
Pix =
jx
L
AE
Pjx = x jx
L
Pix =

Matrix Structural Stiffness 181

Ax E
L

Ax E
L

0
12 EI y
3

L
6 EI y

6 EI y
2

L
4 EI y

L2

12 EI y

6 EI y
2

L
6 EI y

L
2 EI y

L2

Ax E
L
0
0

Ax E
L
0
0

12 EI y
L3
6 EI y
L2
0

12 EI y
L3
6 EI y
L2

6 EI y
2 P
ixix
ix
L

2 EI y iz Piziz
L qiy Miyiy
==

jx Pjxjx

0
jz Pjz
jz
6 EI y q jy M jy
jy
L2

4 EI y
L
0

(4.23)

4.11 ELASTIC MEMBER STIFFNESS, X-Y SYSTEM


The X-Y coplanar coordinate system differs from the X-Z system in a few
ways. The derivation of elastic member stiffness in the X-Y system will
be derived in the two following examples using the area moment method
and conjugate beam method.
Example4.17 qiz stiffness
Derive the qiz stiffness using the area moment method for a linear member.
A free-body diagram is shown in Figure4.30 with an imposed rotation of 1 unit on the i-end of the member. The moments are assumed in the
positive z direction using the right-hand rule and the Cartesian right-hand
coordinate system. The forces are shown consistent with the deformation.
The moment diagram divided by EI is shown for the reaction forces on the
i-end of the member.
Piy

Pjy
Mjz

Miz

iz
Miz
EIz
PiyL
EIz

Figure4.30. Example4.17 qiz stiffness.

182NUMERICAL STRUCTURAL ANALYSIS

Since both ends of the beam are fixed for translation, the tangential
deviation of a point at the i-end from a tangent to the curve on the j-end is
zero. This is equal to the moment of the area of the M/EI diagram about
the point at the i-end.
j

tij = 0 =
i

2
M L L Piy L 2 L
M
xi dx = iz +

EI
EI z 2 2 EI z 3

3M i z
Piy =
2L
The change in rotation from the i-end to the j-end is equal to the negative
of the implied rotation. This is the area under the M/EI diagram between
those points.
qij = 0 qiz = qiz =

2
M iz L Piy L
+
EI z
2 EI z

Substituting the first equation for Piy into the qiz equation results in one of
the stiffness terms. The second term is found by substitution of the first
stiffness term back into the Piy equation.
qiz =

M iz L
4 EI z

4 EI z
qiz
L
6 EI
Piy = 2 z qiz
L

M iz =

(4.24)
(4.25)

Example4.18 Diy stiffness


Derive the Diy stiffness using the conjugate beam method for a linear
member.
A free-body diagram is shown in Figure4.31 with an imposed deflection of one unit on the i-end of the member. The moments are assumed
in the positive z direction using the right-hand rule and the Cartesian
right-hand coordinate system. The forces are shown consistent with the
deformation. The moment diagram divided by EI is shown for the reaction
forces on the i-end of the member.

Matrix Structural Stiffness 183


Piy
Pjy
iy

Mjz
Miz
Miz
EIz
PiyL
EIz

Figure4.31. Example4.18 iy stiffness.

The conjugate beam can be constructed for the two basic assumptions. The shear in the conjugate is equal to the slope of the real beam, and
the moment of the conjugate is equal to the deflection of the real beam.
Viconj = qireal = 0
M iconj = ireal = iy
Since moment in the conjugate exists but shear is zero, the conjugate beam
is slotted in the y direction on the i-end. This is a connection that is free to
move vertically, but restrained from rotation.
V jconj = q jreal = 0
M jconj = jreal = 0
Since both the shear and the moment in the conjugate do not exist, the conjugate beam is free on the j-end. The resulting conjugate beam is shown
in Figure4.32.
The conjugate beam method can be applied to find the reactions at the
i-end of the conjugate beam, which are equal to deformations at the i-end
of the real beam. The load from Figure4.31 is applied to the conjugate
beam in Figure4.32.
Viconj = qireal = 0 =

2
M iz L Piy L
+
EI z
2 EI z

2
M L L Piy L 2 L
M iconj = ireal = iy = iz +

EI z 2 2 EI z 3

184NUMERICAL STRUCTURAL ANALYSIS

Figure4.32. Example4.18 iy stiffness.

Solving the first equation for Piy in terms of Miz and then substituting into
the second equation, the stiffness value can be found.
2 M iz
L
M L2
iy = iz
6 EI z
Piy =

M iz =

6 EI z
iy
L2

(4.26)

This can be substituted back into the equation for Piy to obtain the last
stiffness value.
Piy =

12 EI z
iy
L3

(4.27)

The four terms given in Equations 4.24 through 4.27 are the flexural stiffness terms for the forces at the i-end due to motions at the i-end. This is
denoted as stiffness matrix [Kii] in Equation 4.28. The stiffness equation
and matrix form of this are as follows:

[ K ][ d ] = [ F ]
ii

12 EI z
L3

6 EI z
L2

6 EI z
L2 iy Piy
=

4 EI z iz M iz
L
12 EI z 6 EI z
L3
L2

[ Kii ] = 6 EI
4 EI z

z
L2
L

(4.28)

The transmission matrix derived in Section 4.3 for the X-Y system can be
used to find the forces at the j-end.
1

[T ] = L

0
1

Matrix Structural Stiffness 185

From statics equilibrium, the forces at the i-end transmitted to the j-end
plus the forces at the j-end are equal to zero. The forces at the j-end due to
motions at the i-end are denoted as stiffness matrix [Kji].

[T ][ Kii ] + K ji = 0
K ji = [T ][ K ii ]
12 EI z 6 EI z
12 EI z
2
L3
1
0

1 0 L3

L
K ji =

L 1 6 EI z 4 EI z L 1 6 EI z
L2
L2
L
6 EI
12 EI z
2z
L3
L
K ji =

EI
6
2
EI

z
z
L2
L

6 EI z
L2

4 EI z
L
(4.29)

For a prismatic member it can be shown that the stiffness matrix is symmetrical about the main diagonal. This results in stiffness [Kij] being equal
to the transpose of [Kji], where the forces at the j-end due to motions at the
i-end are denoted as stiffness matrix [Kji].
12 EI z
L3
T
K ij = K ji =
6 EI z
L2
12 EI y 6 EI y
3

L2
L
K ij =
6 EI y 2 EI y
2

6 EI z
12 EI z
2
L3
L
=
2 EI z
6 EI z

L
L2

6 EI z
L2

2 EI z
L
(4.30)

From statics equilibrium, the forces at the i-end transmitted to the j-end
plus the forces at the j-end are equal to zero. The forces at the j-end due to
motions at the j-end are denoted as stiffness matrix [Kjj].

[T ] Kij + K jj = 0
K jj = [T ] K ij
12 EI y
3
1 0
L

K jj =

L 1 6 EI y
2
L

6 EI z
12 EI z

6 EI y
12 EI y
3
2
L 1 0
L
=

2 EI y L 1 6 EI y

2
L
L

6 EI y

L2
2 EI y

[T ] Kij + K jj = 0
K jj = [T ] K ij

186NUMERICAL STRUCTURAL ANALYSIS

12 EI y
3
1 0
L

K jj =

6
L
1

EI

y
2
L

6 EI
12 EI z
2z
L3
L
K jj =

4 EI z
6 EI z
L2
L

6 EI y
12 EI y
3
2
L 1 0
L
=
2 EI y L 1 6 EI y

2
L
L

6 EI y

L2
2 EI y

L
(4.31)

If we add the axial stiffness terms that were shown in the previous section,
the entire coplanar X-Y frame stiffness matrix can be found.
Ax E
L

AE
x
L

12 EI z
L3
6 EI z
L2

6 EI z
L2
4 EI z
L

12 EI z
L3
6 EI z
L2

6 EI z
L2
2 EI z
L

Ax E
L
0
0

Ax E
L
0
0

0
12 EI z
L3
6 EI
2z
L

0
12 EI z
L3
6 EI
2z
L

6 EI z
P
L2 ix ixix

iy
2 EI z Piyiy
L qiz Miziz
== P
0 jx jxjx
jy Pjyjy


6 EI z q jz Mjz
2 jz
L
4 EI z

L
0

(4.32)

4.12 ELASTIC MEMBER STIFFNESS, 3-D SYSTEM


By combining Equations 4.23 and 4.32 and adding the torsional stiffness
terms that are derived in most strength of materials textbooks, we can
construct the elastic member stiffness in the three-dimensional (3-D) system Cartesian coordinate system. This will include axial and torsional
stiffness, as well as bending about each orthogonal axes of the member
cross-section.
I xG
qix
L
IG
M jx = x qix
L
I xG
M ix =
qix
L
IG
M jx = x qix
M ix =

I x G Structural Stiffness 187


Matrix

qix
L
IG
M jx = x qix
L
I xG
M ix =
qix
L
IG
M jx = x qix
L
M ix =

AE

Ax E
0
0
0
0
0 x
0
0
0
0
0
L
L

6 EI z
12 EI
6 EI z
0 12 EI z
0
0
0
0 3 z
0
0
0
3
2
2

L
L
L
L

12
6
12
6
E
I
EI
EI
EI
y
y
y
y

0
0

0
0
0
0
0
0
3
2
3
2

L
L
L
L
ix Pix

I xG
I xG
0
0
0
0
0
0
0
0
0
0 iy Piy

L
L
iz Piz

6 EI y
4 EI y
6 EI y
2 EI y

0
0
2
0
0
0
0
0
0 qix M ix
2

L
L
L
L

M
q

6 EI z
4 EI z
6 EI
2 EI z iy iy
0
0
0
0
0 2z
0
0
0
q
2

L
L
L
L iz = M iz
jx Pjx
AE
Ax E

x
0
0
0
0
0
0
0
0
0
0
Pjy
L
jy
L

12 EI
6 EI
12 EI z
6 EI
P
0
0
0 2z 0
0
0
0 2 z jz jz
0 3 z
3

L
L
L
L
q jx
M jx

12 EI y
6 EI y
12 EI y
6 EI y

q
M
jy
0 3
0
0
0
0
0
0 jy
0
2
3
2

L
L
L
L
q jz M jz

I xG
I xG
0
0
0
0
0
0
0
0
0
0
L
L

6 EI y
2 EI y
6 EI y
4 EI y
0
0
0
0
0
0
0
0

L2
L
L2
L

6 EI z
2 EI z
6 EI z
4 EI z
0
0
0
0
0

0
0
0

L2
L
L2
L

(4.33)

4.13GLOBAL JOINT STIFFNESS


The solution of structures using the global joint stiffness method requires
that each component be in the same Cartesian coordinate system. The
main components of the system are the members, loads, and supports. The
main results desired from the solution are the joint deformations, member
forces, and support reactions. The global or joint system is the primary
way to organize all the components. The supports and some of the loads
will be expressed in this system directly. Some of the loads will have to

188NUMERICAL STRUCTURAL ANALYSIS

be transformed into this system. The basic process to set up and solve the
structure using this stiffness is summarized as follows:
1. Find the local member stiffness, [Km], using Equation 4.33 (4.23
or 4.32 for 2-D systems). Rotate the local member stiffness to the
global joint stiffness system, [Kg]. This is shown in Equation 4.34.
K g = [ R ] [ K m ][ R ]
T

(4.34)

2. Assemble all of the members into the global joint stiffness matrix.
This is done by using joint labeling to order the matrix.
3. Determine the global joint loading, [Pg], from all direct loads on
joints, [P & Mg], and loads on members. The member loads are
applied as the opposite of the fixed-end forces and moments,
[FEPMm]. These must be rotated from the local system to the global
system, [R]T.
Pg = P & M g [ FEPM m ][ R ]

(4.35)

4. Solve the general stiffness equation for global displacements, [Dg].


The rows and columns of the matrices corresponding to the deformations restrained by the supports are removed prior to solving the
system of equations.
K g g = Pg

(4.36)

5. Determine the reactions at the support, [Pg], from the global deformations. Any fixed-end forces must be subtracted from the results.
Only the forces at the supports due to the free deformations need
to be found.
Pg = K g g + [ FEPM m ][ R ]

(4.37)

6. Solve for the local member forces and moments, [P & Mm], for each
member separately from the global joint deformations. The global
joint deformations must be rotated into the member system, [R].
The fixed-end forces and moments must be added back to get the
final member end forces.

[ P & M ] = [ K ][ R ]
m

+ [ FEPM m ]

(4.38)

Matrix Structural Stiffness 189

The general set-up for the stiffness method of analysis represents a system of linear equations, where the displacement vector is the unknown.
Except for those designated as supports, there are six unknown joint displacement components for each joint in the structure in a 3-D structure.
Each displacement released at a support is still an unknown displacement
component to the system. There is an equation for each degree of freedom
of the structure. Each non-related component at a support has a displacement that is set identically equal to zero, and as far as the system of equations is concerned, this particular equation may be omitted, along with
any coefficient in the other equations which corresponds to the dropped
displacement.
Sometimes the system solution is handled in six by six blocks of
coefficients or six rows of equations at a time, where each block representing the accumulated stiffness for a joint in the case of the diagonal,
or the carry-over effects from other joints in the case of off-diagonals.
In this case, unless the support joint is fully restrained, its corresponding row of six by six blocks is maintained intact and a number of sufficient sizes to simulate infinite stiffness in the restrained direction
are added to the diagonal of the diagonal block in the master stiffness
matrix.
The building of the global joint stiffness matrix consists of various
stages of operations. First, the member stiffness matrix is defined in its
own system, giving due consideration to member end releases, for each
member in the structure. This will be discussed in Chapter 5. The member
stiffness matrix can be considered as four separate six by six blocks. These
represent the forces at the ends due to the motions at the end and were
discussed in Sections 4.10 and 4.11.
K ii

[K ] = K
m

ji

K ij
K jj

The four separate components of the global member stiffness should be


placed in the global joint stiffness matrix. The diagonal terms will be added
to other terms representing the stiffness of other members connected to
that joint. The off-diagonal terms will simply be placed in their appropriate position. The transformation of the local member system to the global
joint system was shown in Equation 4.34 and is derived as follows starting
with the local stiffness equation:

[ K ][ ] = [ P ]
m

190NUMERICAL STRUCTURAL ANALYSIS

This equation represents the force in the local system due to deformations
in the local system. To go from local to global we multiply by the rotation
transpose, [R]T, on both sides of the equation.

[ K m ][ R ] [ m ] = [ R ] [ Pm ]
T

The right side of the equation now represents the forces in the global system, [Pg]. The left side represents the force in the global system due to
local deformations. The equation needs to be written in terms of the global
deformation, [Dg]. From Section 4.3, the local deformation is the global
deformation multiplied by the rotation matrix, [R].

[ R ]T [ K m ][ R ] g = Pg
Example4.19 Global joint stiffness
Determine the global joint deformations, support reactions, and local
member forces for the pin-connected bracing structure loaded as shown
in Figure4.33.
The area of each member, Ax, is 10 in2 and the modulus of elasticity,
E, is 10,000 ksi. Note that the structure is in the XZ coordinate system.
Since this is a pin connected structure loaded only at the joint, it will
act as a true truss with only axial forces in the members. The stiffness
model will be simplified to only include the axial stiffness components,
AE/L. Furthermore, rotation at the joints will be excluded since there is no
rotational stiffness imparted by the members. Rotation could be included
Z
8k

8k
9k

12k

15'-0"
X

20'-0"

Figure4.33. Example4.19 Global joint stiffness.

Matrix Structural Stiffness 191


Z

4
1

2
4

Figure4.34. Example4.19 Global joint stiffness.

if it is desired to find the rotation of the member ends, but this can easily
be found from the final deformed position of the structure.
A numbering system must be assigned to the joint and members for
easy bookkeeping. Figure 4.34 shows a numbering system for the five
joints and the seven members of the structure. The member numbers are
circled for clarity.
The member stiffness for each member is first found from Equation 4.34
(step 1).
K g = [ R ] [ K m ][ R ]
T

K g = [ b ] [ K m ][ b ]
T

Expand the general local member stiffness to include just the x and z
forces and motions. Also note that rotation transformation of members
will be about the y-axis or b.
AE
0
0 L
cos b sin b

sin b cos b
0
0 0
K g =
0
0
cos b sin b AE


0
sin b cos b L
0
0
cos b sin b
sin b cos b

0
0

0
0

AE
L
0
AE
L
0

0
0
0
0

0
0
0
0
cos b sin b

sin b cos b

To simplify the process, we can multiply the expanded global stiffness


equation and factor out the axial stiffness term.

192NUMERICAL STRUCTURAL ANALYSIS

cos 2 b
sin b cos b

sin 2 b
AE sin b cos b
K g =
L cos 2 b
sin b cos b

sin 2 b
sin b cos b

cos 2 b
sin b cos b
cos 2 b
sin b cos b

sin b cos b

sin 2 b
sin b cos b

sin 2 b

Members 1 and 2 have the same orientation. Selecting the bottom


end as the i-end, the rotation is -90 or 270. The axial stiffness is
555.6 k/in.
0
0
0 555.6
[ K14 ] = [ K35 ] = 0 0

0 555.6

0
0
0 555.6 K11
=
0
0 K 41

0 555.6

K14 K 33
=
K 44 K 53

K 35
K 55

Member 3 does not need rotations since it is already in the global system
orientation. The global stiffness will be the same as the member stiffness.
Selecting the left end as the i-end, the rotation is 0. The axial stiffness is
416.7 k/in.
416.7
0
K
=
[ 45 ] 416.7

0 416.7
0
0
0 416.7
0
0

0
0 K 44
=
0 K 54

K 45
K 55

Members 4 and 5 have the same orientation. Selecting the bottom-left end
as the i-end, the rotation is -36.87 or 323.13. Instead of using angles, in
this case it is easier to use trigonometry directly. The cosine is 0.8 and the
sine is -0.6. The axial stiffness is 666.7 k/in.
426.7 320.0 426.7 320.0
320.0
240.0 320.0 240.0 K
[ K12 ] = [ K 25 ] = 426.7 320.0 426.7 320.0 = K11
21

240.0
320.0 240.0 320.0
K 22
=
K 52

K 25
K 55

K12
K 22

Matrix Structural Stiffness 193

Members 6 and 7 have the same orientation. Selecting the top-left end as
the i-end, the rotation is 36.87 or -323.13. Instead of using angles, in
this case it is easier to use trigonometry directly. The cosine is 0.8 and the
sine is 0.6. The axial stiffness is 666.7 k/in.
426.7 320.0 426.7 320.0
320.0 240.0 320.0 240.0
K
[ K 23 ] = [ K 42 ] = 426.7 320.0 426.7 320.0 = K 22
32

320
.
0

240
.
0

320
.
0
240
.
0

K 44
=
K 24

K 23
K 33

K 42
K 22

The global joint stiffness matrix can be assembled using each of the members contributions (step 2).
K11
K
21
K g = 0

K 41
0
427 320
320 796

427 320

320 240
0
0
K g =
0
0
0
0

0
566

0
0
0
0

427
320
1707
0
427
320
427
320
427
320

K12
K 22
K 32
K 42
K 52

0
K 23
K 33
0
K 53

K14
K 24
0
K 44
K 54

0
K 25
K 35

K 45
K 55

0
0
0
0
0
0
320
0
0
0
0
0
240
566
0
427 320 427 320 427 320

960 320 240 320 240 320 240


0
0
0
0
320
427 320

0
0
0
240 320 796
566
320
0
0
843 320 417
0

0
0
240
320 796
0
0

0
0
0
843 320
320
417
240
0
566
0
0
320
795

The global joint loading can be determined from Equation 4.35 directly
since all of the applied loads are at the joints and in the global system
(step 3). Note that there are no loads applied directly to the members, so
there are no fixed-end forces and moments. The load matrix is in units of
kips (k).

194NUMERICAL STRUCTURAL ANALYSIS

Pg = P & M g [ FEPM m ][ R ]

P1x 0
P
1y 0
P2 x 0

P2 y 0
P3 x 0
= =
P3 y 0
P 12
4x
P4 y 8

P5 x 9
P6 y 8

Using any of the methods for solving non-homogenous linear algebraic


equations, the global deformations can be found from the global stiffness
Equation 4.36 (step 4). The rows and columns corresponding to the support constraint degrees of freedom must be deleted prior to the solution.
This would be both x and y at joints 1 and 3. The solution for the deformations will be in inches.
1

g = K g Pg
1

0
427 320 427 320 0 0.0246
2 x 1707

960 320 240 320 240 0 0.0057
2y 0
4 x 427 320
843 320 4117
0 12 0.0595
=

320

240

320
796
0
0 8 0.0022
4y
5 x 427 320 417
0
843 320 9 0.0602

0
0
320
795 8 0.0261
6 y 320 240
The reactions at the support can be found using the solution of the global
deformation with Equation 4.37 (step 5). Only the terms in the rows corresponding to the restrained degrees of freedom and in the columns of
the unrestrained degrees of freedom need to be included. Since there are
no applied loads on the members, the fixed-end forces and moments are
omitted. In addition, there are no applied loads at the support locations so
the global applied forces and moments are omitted. The reaction forces
are in kips (k).

[ P ] = K g g P & M g [ FEPM m ][ R ]T
0
0
427 320 0
320 240 0 566 0
[ P ] = K g g = 427 320 0 0 0

0.0246
0 0.0057
0 0.0595

0 0.0022

Matrix Structural Stiffness 195

[ P ] = K g g P & M g [ FEPM m ][ R ]T
427 320
320 240
[ P ] = K g g = 427 320

320 240

0
0
0 566
0
0
0
0

0
0
0
0

0.0246
0 0.0057
0 0.0595

0 0.0022

566 0.0602

0.0261

8.67
7.75

=
12.33

23.75
The final step is finding the member forces for each of the members using
Equation 4.35 (step 6). Since there are no applied forces on the members, the fixed-end forces and moments can be omitted. Note that the local
member stiffness matrix is used here and not the global matrix. The member force will be in kips. If the i-end is positive, the member is in compression and if it is negative the member is in tension.

[ P & M ] = [ K ][ R ]
[ P & M ] = [ K ][ b ]
m

+ [ FEPM m ]

For member 1, the deformations at joints 1 and 4 are used.


0
0
555.6
0
555.6
0

555.6
0
555.6

0
0
0

0 0
0 1
0 0

0 0

1 0
0 0
0 0
0 1

0 0 1.24
0 0 0
=
1 0.0595 1.24

0 0.0022 0

For member 2, the deformations at joints 3 and 5 are used.


0
0
555.6
0

555
.
6
0

555.6
0
555.6

0
0
0

0 0
0 1
0 0

0 0

1 0
0 0
0 0
0 1

0 0 14.5
0 0 0
=
1 0.0602 14.5

0 0.0261 0

196NUMERICAL STRUCTURAL ANALYSIS

For member 3, the deformations at joints 4 and 5 are used.


416.7
0

416.7

0 416.7
0
0
0 416.7
0
0

0 1
0 0
0 0

0 0

0
1
0
0

0
0
1
0

0 0.0595 0.3326
0 0.0022 0
=
0 0.0602 0.326

1 0.0261 0

For member 4, the deformations at joints 1 and 2 are used.


666.7
0

666.7

0 666.7
0
0
0 666.7
0
0

0 0.8 0.6
0
0 0 10.8

0
0 0 0
0 0.6 0.8
=
0
0.8 0.6 0.0246 10.8
0 0

0 0.6 0.8 0.0057 0


0 0

For member 5, the deformations at joints 2 and 5 are used.


666.7
0

666.7

0 666.7
0
0
0 666.7
0
0

0 0.8 0.6
0
0 0.0246 10.8

0
0 0.0057 0
0 0.6 0.8
=
0
0.8 0.6 0.0602 10.8
0 0

0 0.6 0.8 0.0261 0


0 0

For member 6, the deformations at joints 2 and 3 are used.


666.7
0

666.7

0 666.7
0
0
0 666.7
0
0

0 0.8 0.6
0
0 0.0246 15.4

0
0 0.0057 0
0 0.6 0.8
=
0
0.8 0.6 0 15.4
0 0

0 0.6 0.8 0 0
0 0

For member 7, the deformations at joints 4 and 2 are used.


666.7
0

666.7

0 666.7
0
0
0 666.7
0
0

0 0.8 0.6
0
0 0.0595 15.4

0
0 0.0024 0
0 0.6 0.8
=
0
0.8 0.6 0.0246 15.4
0 0

0 0.6 0.8 0.0057 0


0 0

Matrix Structural Stiffness 197

Example4.20 Global joint stiffness


Determine the global joint deformations, support reactions, and local
member forces for the rigidly connected frame structure loaded as shown
in Figure4.35.
The area of each member, Ax, is 10 in2, the moment of inertia, Iz, is
1000 in4, and the modulus of elasticity, E, is 10,000 ksi. Note the structure
is in the XY coordinate system and the numbering system is similar to
Example4.19.
Y

3 k/ft

18k
2
16'-0"

24'-0"

Figure4.35. Example4.20 Global joint stiffness.

The member stiffness for each member is first found from


Equation4.34 (step 1).
K g = [ R ] [ K m ][ R ]
T

K g = [a ] [ K m ][a ]
T

Using the local member stiffness for the XY system from Section 4.11
which has translation in x and y direction and rotation about the z direction, the rotation transformation of members will be about the z-axis or a.

[a ]T

cos sin
sin cos

0
0
=
0
0
0
0

0
0

0
0
0
0
0
0
1
0
0
0 cos sin
0 sin cos
0
0
0

0
0
0

0
0

198NUMERICAL STRUCTURAL ANALYSIS

AE
L

[ K m ] = AE

12 EI z
L3
6 EI z
L2

6 EI z
L2
4 EI z
L

12 EI z
L3
6 EI z
L2

0
0

6 EI z
L2
2 EI z
L

cosa sina
sina cosa

0
0
[a ] = 0
0

0
0

0
0

AE
L

AE
L
0
0

0
12 EI z
L3
6 EI
2z
L

0
12 EI z
L3
6 EI
2z
L

0
0
0
0
0
0
1
0
0
0 cosa sina
0 sina cosa
0
0
0

6 EI z
L2

2 EI z
L

6 EI z
2
L
4 EI z

L
0

0
0
0

0
0

Members 1 and 3 have the same orientation. Selecting the bottom end as
the i-end, the rotation is 90. Table4.3 contains the local member stiffness
matrix, the rotation matrices, a and aT, and the global member stiffness
matrix.
K11

[K ] = [K ] = K
12

43

21

K12 K 44
=
K 22 K 34

K 43
K 33

Member 2 is already in the global system and does not need rotation.
Selecting the left end as the i-end, the rotation is 0. Table4.4 contains
the local member stiffness matrix, the rotation matrices, a and aT, and the
global member stiffness matrix.
K 22

[K ] = K
23

32

K 23
K 33

Matrix Structural Stiffness 199

Table4.3. Example4.20 Global joint stiffness


1510.4
0
0
-1510.4
0
0

0
49.17
4720
0
-49.17
4720

0
4720
604166.67
0
-4720
302083.33

0
-1
0
0
0
0

1
0
0
0
0
0

0
0
1
0
0
0

0
1
0
0
0
0

-1
0
0
0
0
0

0
0
1
0
0
0

49.2
0

0
1510

-4720
0

-4720
-49.2
0
-4720

0
0
-1510
0

604166.67
4720
0
302083.33

KL
-1510.4
0
0
1510.4
0
0

0
-49.17
-4720
0
49.17
-4720

0
4720
302083.33
0
-4720
604166.67

0
0
0
0
-1
0

0
0
0
1
0
0

0
0
0
0
0
1

0
0
0
0
1
0

0
0
0
-1
0
0

0
0
0
0
0
1

KG
-49.2
0

0
-1510

-4720
0

4720
49.2
0
4720

0
0
1510
0

302083.33
4720
0
604166.67

aT

The global joint stiffness matrix can be assembled using each of the members
contributions (step 2). Table4.5 contains the global joint stiffness matrix.
The global joint loading is determined from Equation 4.35. In this
case, member 2 is loaded with a uniformly distributed load. The fixed-end
forces and moments due to the load must be calculated. Normally, the
fixed-end forces and moments are rotated into the global system before
they are placed in the global joint loading, but in this case, the member is
already in the global system and no rotation is necessary (step 3). The load
matrix is in units of kips and inches (k-in).

200NUMERICAL STRUCTURAL ANALYSIS

Table4.4. Example4.20 Global joint stiffness


KL
1006.9

-1006.9

14.57

2098

-14.57

2098

2098

402777.78

-2098

201388.89

-1006.9

1006.9

-14.57

-2098

14.57

-2098

2098

201388.89

-2098

402777.78

a
1

aT
1

1006.9

-1006.9

15

2098

-15

2098

2098

402777.78

-2098

201388.89

-1006.9

1006.9

KG

-15

-2098

15

-2098

2098

201388.89

-2098

402777.78

Table 4.5. Example 4.20 Global joint stiffness


K11
K
K g = 21
0

K12
K 22
K 32

0
K 23
K 33

K 43

0
0
K 34

K 44

Matrix Structural Stiffness 201

FEP23 =

wL 0.25k /in ( 288in)


=
= 36k
2
2

wL2 0.25k /in ( 288in)


=
= 1728k in
12
12
wL 0.25k /in ( 288in)
FEP32 =
= 36k
=
2
2
2

FEM 23 =

0.25k /in ( 288in)


wL2
=
= 1728k in
12
12
2

FEM 32 =

The 18 k lateral load is placed directly on joint 2 in the y direction.

Pg = P & M g [ FEPM m ][ R ]

P1x 0
P 0
1y

M 1z 0

P2 x 18
P2 y 36

M 2 z 1728
=
=
P 0
3x

P3 y 36

M 3 z 1728
P4 x 0

P4 y 0
M 0

4z

The global deformations can be found from the global stiffness Equation 4.36 (step 4). The rows and columns corresponding to the support
constraint degrees of freedom must be deleted prior to the solution. This
would be all three motions at 1 and 4. The resulting matrix is just joints 2
and 3. This is shown in Table4.6 along with the reduced load. The solution
for the deformations will be in inches and radians.
2 x 0.3040

2 y 0.0207
q 0.0034
1
g = K g Pg = 2 z =

3 x 0.2852
3 y 0.0270

q 3 z 0.0010

202NUMERICAL STRUCTURAL ANALYSIS

Table4.6. Example4.20 Global joint stiffness


KG

P-FEPM

1056

4720

-1007

18

1525

2098

-15

2098

-36

4720

2098

1006944

-2098

201389

-1728

-1007

1056

4720

-15

-2098

1525

-2098

-36

2098

201389

4720

-2098

1006944

1728

The reactions at the supports can be found using the solution of the global
deformation with Equation 4.37 (step 5). Only the terms in the rows corresponding to the restrained degrees of freedom and in the columns of the
unrestrained degrees of freedom need to be included. Table4.7 shows the
appropriate stiffness terms and deformations needed to find the reactions.
Since there were no fixed-end forces and moments at the support joints,
the solution is complete. The reaction forces are in kips and inches (k-in).

[ P ] = K g g P & M g [ FEPM m ][ R ]T
[ P ] = K g g
Table4.7. Example4.20 Global joint stiffness
KG2
-49

DG

-4720

0.3040

P
0.92

-1510

-0.0207

31.22

4720

302083

-0.0034

419.18

-49

-4720

0.2852

-18.92

-1510

-0.0270

40.78

4720

302083

0.0010

1659.74

The final step is finding the member forces for each of the members using
Equation 4.35 (step 6). The member force will be in kips and inches.
The local member stiffness matrix and the rotation matrix were shown
in step1 and are omitted here. The sign convention for the X-Y system
applies when interpreting the final-end forces and moments.

[ P & M ] = [ K ][ R ]
[ P & M ] = [ K ][a ]
m

+ [ FEPM m ]

Matrix Structural Stiffness 203

For member 1, the deformations at joints 1 and 2 are used. Table4.8 contains the final member-end forces in the local system along with global
deformations used to find those end forces.
Table4.8. Example4.20 Global joint stiffness
PL

DG
0

Dx1

31.22

kips

Dy1

-0.92

kips

z1

419.18

kip-in

0.3040

Dx2

-31.22

kips

-0.0207

Dy2

0.92

kips

-0.0034

z2

-596.37

kip-in

For member 2, the deformations at joints 2 and 3 are used. Table4.9 contains the final member-end forces in the local system along with global
deformations used to find those end forces.
Table4.9. Example4.20 Global joint stiffness
DG

PL

0.3040

Dx2

18.92

-0.0207

kips

Dy2

-4.78

kips

-0.0034

qz2

-1131.63

kip-in

0.2852

Dx3

-18.92

kips

-0.0270

Dy3

4.78

kips

0.0010

qz3

-245.46

kip-in

For member 3, the deformations at joints 4 and 3 are used. Table4.10


contains the final member-end forces in the local system along with global
deformations used to find those end forces.
Table4.10. Example4.20 Global joint stiffness
PL

DG
0

Dx4

40.78

kips

Dy4

18.92

kips

qz4

1659.74

0.2852

Dx3

-40.78

kips

-0.0270

Dy3

-18.92

kips

qz3

1973.46

0.0010

kip-in

kip-in

204NUMERICAL STRUCTURAL ANALYSIS

REFERENCES
Castigliano, A. 1879. Theorie de lEquilibre des Systemes Elastiques et ses Applications. Turin, Italy.
Cross, H. 1930. Analysis of Continuous Frames by Distributing Fixed-End
Moments. Proceedings of the American Society of Civil Engineers (ASCE),
Cleveland, Ohio.
Leibniz, G. 1695. Specimen Dynamicum.
Lewis, D.C.; Review: Frazer, R.A., Duncan, W.J., and Collar, A.R. 1939. Elementary Matrices and some Applications to Dynamics and Differential Equations.
New York, New York: Bulletin of the American Mathematical Society 45.
Maney, G.A. 1915. Studies in Engineering. Minneapolis, Minnesota: University
of Minnesota
Mller-Breslau, H. 1875. Elementary Handbook of Strength of Materials.
Timoshenko, S.P. 1953. History of Strength of Materials. New York, New York:
McGraw-Hill.

CHAPTER 5

Advanced Structural
Stiffness
In this chapter, concepts learned in previous chapters are expanded and
applied to advanced structural stiffness. This method is applied in computers to solve complex structures that are statically determinate or statically
indeterminate. Matrices will be utilized to determine internal member
forces and displacements within a structure. Small pieces of the structure are analyzed and then compiled into a larger matrix in order to view
the structure as a whole. This procedure is the basis for finite element
analyses.

5.1 MEMBER END RELEASES, X-Z SYSTEM


Joint stiffness is expressed in the master matrix for a structure, but two
situations exist that may cause them to vary. First, a joint that is being utilized as a support may be released. For example, the support becomes slotted or pinned. When looking at a joint release, the joint is fully designed as
a support before the release. This creates the reaction components. A support release is in the global system and is handled in the reduction of the
joint stiffness matrix. Second, a member can be physically released from a
joint for one or more of the six possible end displacements. When a member release occurs, the member is released in some direction and the stiffness contribution that member was making to the joint changes or goes to
zero. When a member is released in the local system, this release changes
the member stiffness matrix. Some of these released stiffness matrices will
be derived in the following examples. The order of the motions and forces
on the members end are given by the deflections and rotations at the i-end
followed by the deflections and rotations at the j-end. The following is a
full list of stiffness values simplified from Equation 4.33:

206NUMERICAL STRUCTURAL ANALYSIS

k1,1
0

0
0

0
k
7 ,1
0

0
0

0
0

0
k2, 2
0
0
0
k6 , 2
0
k8, 2
0
0
0
k12, 2

0 0
0 0
k3 , 3 0
0 k4, 4
k5 , 3 0
0 0
0 0
0 0
k9 , 3 0
0 k10, 4
k11,3 0
0 0

0
0
k3 , 5
0
k5 , 5
0
0
0
k9 , 5
0
k11,5
0

0
k2, 6
0
0
0
k6 , 6
0
k8, 6
0
0
0
k12, 6

k1, 7
0
0
0
0
0
k7 , 7
0
0
0
0
0

0
k 2 ,8
0
0
0
k6 ,8
0
k8,8
0
0
0
k12,8

0
0
0
0
k3, 9 0
0 k4,10
k5 , 9 0
0
0
0
0
0
0
k9 , 9 0
0 k10,10
k11,9 0
0
0

0
0
k3,11
0
k5,11
0
0
0
k9,11
0
k11,11
0

0 ix PPixix
k2,12 iy PPiyiy

0 iz PPiziz


0 qix Mixix
0 qiy Miyiy


k6,12 qiz Miziz
=
=
0 jx PPjxjx


k8,12 jy PPjyjy


0 jz PPjzjz
0 q jx M jxjx


0 q jy M jyjy
k12,12 q jz M jzjz

(5.1)
Example5.1 iz end release
Derive the local member stiffness for a Diz member end release using the
conjugate beam method.
A free-body diagram of the released beam is shown in Figure5.1.
Since the beam is allowed to move at the i-end in the z direction, the
reaction Piz is equal to zero. The loaded conjugate beam is also shown in
Figure5.1. Note that the shear in the conjugate beam is equal to the rotation in the real beam and the moment in the conjugate beam is equal to the
deflection in the real beam.
Miy
Piz=0

Miy

iy

Miy
EIy

Pjz

Miy
EIy

iz=0

iz=0

iy

iy=0

Figure5.1. Example5.1 iz end release.

If a motion Diz is imposed, there is no resistance and therefore no


forces. The resulting forces are the stiffness values due to the motion. The
following are forces and stiffness due to Diz:

Advanced Structural Stiffness 207

Piz = M iy = Pjz = M jy = 0
k3,3 = k5,3 = k9,3 = k11,3 = 0
If a motion Djz is imposed, there is no resistance and therefore no forces.
The resulting forces are the stiffness values due to the motion. The following are forces and stiffness due to Djz:
Piz = M iy = Pjz = M jy = 0
k3,9 = k5,9 = k9,9 = k11,9 = 0
If a motion qiy is imposed, there is resistance and therefore forces. The
resulting forces are derived using conjugate beam.

Fz = qiy
M iy =
k5 , 5 =

EI y
L
EI y
L

M iy L
EI y

=0

qiy
qiy

From statics on the real beam:

M y = M iy + M jy = 0
M jy = M iy =
k11,5 =

EI y

L
Piz = Pjz = 0

EI y
L

qiy

qiy

k3 , 5 = k9 , 5 = 0
From symmetry of the stiffness matrix, the following terms can be
found:

k5,11 = k11,5
k5,11 =

EI y
L

q jy

208NUMERICAL STRUCTURAL ANALYSIS

From statics on the real beam considering qjy:

M iy = M jy =
EI y

EI y
L

q jy

q jy
L
Piz = Pjz = 0

k11,11 =

k3,11 = k9,11 = 0
The resulting stiffness matrix is shown in Equation 5.2 with only the
affected terms replaced with the new values.
k1,1
0

k7 ,1
0

0
k2, 2
0
0

0
0
0
0

0
0
0
k4, 4

0 0 0
k6 , 2
0
k8, 2
0
0

0
0
0
0
0

0
0
0
0
k10, 4

0 0 0
k12, 2 0 0

0
0
0 k2, 6
0
0
0
0
EI y
0
L
0 k6 , 6
0
0
0 k8, 6
0
0
0
0
EI y

0
L
0 k12, 6

k1, 7 0
0 k 2 ,8
0 0
0 0
0
0
k7 , 7
0
0
0
0

0
0
0
0

0
0
0
k4,10

0 0 0
k6 ,8
0
k8,8
0
0

0
0
0
0
0

0
0
0
0
k10,10

0 0 0

0 k12,8 0 0

0
0
0 k2,12 ix

0
0 iy

0
0 iz

qix
EI y

0
L
qiy
0 k6,12 qiz

0
0 jx

0 k8,12 jy

0 jz
0

0
0 q jx

q jy
EI y
0
L
q jz
0 k12,12

Pix
P
iy
Piz

M ix
M
iy
M iz
=
P
jx
Pjy

Pjz
M jx

M jy
M
jz

(5.2)
Example5.2 qiy end release
Derive the local member stiffness for a qiy member end release using the
conjugate beam method.
A free-body diagram of the released beam is shown in Figure5.2.
Since the beam is allowed to rotate at the i-end in the y direction, the
reaction Miy is equal to zero. The loaded conjugate beam is also shown in
Figure5.2. Note that the shear in the conjugate beam is equal to the rotation in the real beam and the moment in the conjugate beam is equal to the
deflection in the real beam.
If a motion Diz is imposed, there is resistance and therefore forces. The
resulting forces are derived using conjugate beam.

Advanced Structural Stiffness 209

Miy=0
Mjy

iz
Piz

Pjz
PizL
EIy

iz

jz=0

iy

jy=0

Figure5.2. Example5.2 qiy end release.

M jy = iz +
Piz =
k3 , 3 =

3EI y
L3
3EI y
L3

Piz L L 2 L
=0
EI y 2 3

iz
iz

From statics on the real beam:


M jy = Piz L + M jy = 0
M jy = Piz L =
k11,3 =

3EI y

3EI y

L2

iz

iz
L2
Fz = Piz + Pjz = 0
Pjz = Piz =
k9 , 3 =
M iy = 0
k5 , 3 = 0

3EI y
L3

3EI y
L3

iz

iz

210NUMERICAL STRUCTURAL ANALYSIS

If a motion Djz is imposed, there is resistance and therefore forces. The


free-body diagram of the deflected shape is shown in Figure 5.3. The
conjugate beam will be the same as shown in Figure5.2. The resulting
forces are derived using conjugate beam. Note that the direction of Piz in
Figure5.3 will be the opposite of that in Figure5.2, so all the values will
be the opposite of those from Diz.
Mjy
Miy=0

jz

Piz

Pjz

Figure5.3. Example5.2 qiy end release.

Piz =

3EI y

k3 , 9 =

L3
3EI y
L3

jz
jz

M jy = Piz L =
k11,9 =

3EI y
L2

Pjz = Piz =
k9 , 9 =

3EI y
L3

3EI y
L2

jz

jz
3EI y
L3

jz

jz

If a motion qiy is imposed, there is no resistance and therefore no forces.


The resulting forces are the stiffness values due to the motion. The following are forces and stiffness due to Diz:
Piz = M iy = Pjz = M jy = 0
k3,5 = k5,5 = k9,5 = k11,5 = 0
If a motion qjy is imposed, there is resistance and therefore forces. The
free-body diagram of the deflected shape is shown in Figure 5.4. The

Advanced Structural Stiffness 211

Mjy

Miy=0

jy

Piz

Pjz

Mjy
EIy

iz=0

jz=0

iy

jy

Figure5.4. Example5.2 qiy end release.

conjugate beam is also shown in Figure5.4. The resulting forces are


derived using conjugate beam.
M iy = q jy L
M jy =
k11,11 =

3EI y
L
3EI y
L

M jy L 2 L
=0
EI y 2 3

q jy
q jy

From statics on the real beam:

M jy = Piz L + M jy = 0
Piz =

M jy

3EI y

L
3EI y

L2

q jy
L2
Fz = Piz + Pjz = 0

k3,11 =

Pjz = Piz =
k9,11 =

3EI y

M iy = 0
k5,11 = 0

L2

3EI y
L2

q jy

q jy

q jy

212NUMERICAL STRUCTURAL ANALYSIS

The resulting stiffness matrix is shown in Equation 5.3 with only the
affected terms replaced with the new values.
k1,1
0

0
0

k7 ,1
0

0
k 2 ,2
0
0
0
k 6 ,2
0
k8,2
0
0
0
k12,2

0
0
3EI y

0 0 0 k1,7 0
0 0 k2,6 0 k2,8

0
L3
k 4 ,4
0
0
0
0
0
0
0
0
0
3EI y
3
0
L
k10,4
0
3EI y
0
2
L
0
0

0 0
0
0
0
0
0

L3
0
0
0
0
0
3EI y

0 0 0
0 0 0
k6,6 0 k6,8
0 k7 ,7 0
k8,6 0 k8,8

0 0

0 0

0 0

0
0
3EI y

0
0
0
k4,10
0
0
0
0

0
L3
k10,10
0
3EI y
0
L2
0
0

0 k12,6 0 k12,8

0
0
0
k2,12
P
3EI y
0 ix ix
L2
iy Piy
0
0 iz Piz

0
0 q ix M ix
0
k6,12 q iy M iy

0
0 q iz = M iz


0
k8,12 jx Pjx
jy Pjy
3EI y
0
jz Pjz
L2
q M
0
0 jx jx
q jy M jy
3EI y

0 q jz M jz
L

k12,12
0

(5.3)
The member stiffness for releasing Djz and qjy can be derived in a similar manner to Diz and qiy. The resulting stiffness matrices are shown in
Equations 5.4 and 5.5 with only the affected terms replaced with the new
values.
k1,1
0

k7 ,1
0

0
k 2 ,2
0
0

0
0
0
0

0
0
0
k 4 ,4

0 0 0
k 6 ,2
0
k8,2
0
0

0
0
0
0
0

0
0
0
0
k10,4

0 0 0
k12,2 0 0

0
0
0
0
EI y

0 k1,7 0
k2,6 0 k2,8
0 0 0
0 0 0

0
L
0 k6,6
0
0
0 k8,6
0
0
0
0
EI y
0

L
0 k12,6

0 0

0 k6,8
k7 ,7 0
0 k8,8
0 0
0 0
0

0 0
0 0
0 0
0 k4,10
0

0 0
0 0
0 0
0 0
0 k10,10

0 0

0 k12,8 0

0
0
0 k2,12 ix Pix

0
0 iy Piy

P
0
0 iz
iz

q
EI y
ix M ix

0
M
L
qiy
iy

0 k6,12 qiz M iz

=
0
0 jx Pjx

0 k8,12 jy Pjy

0
0 jz Pjz

M
0
0 q jx
jx

EI y
q jy M
0 jy
L
q jz M jz
0 k12,12


(5.4)

Advanced Structural Stiffness 213

k1,1
0

0
k
7 ,1
0

0
0

0
k 2 ,2
0
0
0
k 6 ,2
0
k8,2
0
0
0
k12,2

0
0
3EI y

0
0

0
L3
k 4 ,4
0
3EI y
0
2
L
0
0
0
0
0
0
3EI y
3
0
L
k10,4
0
0
0
0
0

0
0
0
k2,6
3EI y
2
0
L
0
0
3EI y
0
L
k6,6
0
0
0
0
k8,66
3EI y
0
L2
0
0
0
0
0
k12,6

k1,7 0
0 k2,8
0

0 k6,8
k7 ,7 0
0 k8,8
0

0 0
0 0
0 k12,8

0
0
3EI y

0
0
0

L3
0
3EI y

k4,10

L2
0
0
0
3EI y

0
0
0

L3
0
0
0

0
k10,10
0
0

0 0
0 k2,12

0 0

0 0

0 0

0 k6,12
0 0

0 k8,12

0 0

0 0
0 0

0 k12,12

ix Pix
P
iy iy
iz Piz


q ix M ix
q M
iy iy
q iz M iz
= P
jx jx
jy Pjy


jz Pjz
q M
jx jx
q jy M jy
q M
jz jz

(5.5)
When more than one of the four degrees of freedom is released on a flexural member, two conditions may exist. The first is that the member will
provide no joint stiffness. This occurs when both deflection and rotation
at either end are released, when rotation is released at both ends, or when
deflection is released at one end and rotation is released at the other end.
The stiffness matrix for this condition is shown in Equation 5.6. The
resulting beam is either a cantilever beam, a pinned-pinned beam, or a
pinned-slotted beam.
k1,1
0

0
0

0
k
7 ,1
0

0
0

0
0

0
k2, 2
0
0
0
k6 , 2
0
k8, 2
0
0
0
k12, 2

0 0
0 0
0 0
0 k4, 4
0 0
0 0
0 0
0 0
0 0
0 k10, 4
0 0
0 0

0
0
0
0
0
0
0
0
0
0
0
0

0 k1, 7
k2, 6 0
0
0
0
0
0
0
k6 , 6 0
0 k7 , 7
k8, 6 0
0
0
0
0
0
0
k12, 6 0

0
k 2 ,8
0
0
0
k6 ,8
0
k8,8
0
0
0
k12,8

0 0
0 0
0 0
0 k4,10
0 0
0 0
0 0
0 0
0 0
0 k10,10
0 0
0 0

0
0
0
0
0
0
0
0
0
0
0
0

0 ix Pix
k2,12 iy Piy
0 iz Piz


0 qix M ix
0 qiy M iy


k6,12 qiz M iz
=
0 jx Pjx


k8,12 jy Pjy


0 jz Pjz
0 q jx M jx


0 q jy M jy
k12,12 q jz M jz
(5.6)

214NUMERICAL STRUCTURAL ANALYSIS

The second condition is an unstablebeam. This occurs when both ends


are released for deflection or more than two of the four degrees of freedom are released. Table 5.1 summarizes all the flexural stiffness conditions for the X-Z system. In this table, 1 indicates that the degree of
freedom is released and 0 indicates that the degree of freedom is not
released.
Table5.1. Release codesX-Z system
Diz

qiy

Djz

qjy

Equation

0
1
0
0
0
1
1
1
0
0
0
0
1
1
1
1

0
0
1
0
0
1
0
0
1
1
0
1
0
1
1
1

0
0
0
1
0
0
1
0
1
0
1
1
1
0
1
1

0
0
0
0
1
0
0
1
0
1
1
1
1
1
0
1

4.33
5.2
5.3
5.4
5.5
5.6
Unstable
5.6
5.6
5.6
5.6
Unstable
Unstable
Unstable
Unstable
Unstable

5.2 MEMBER END RELEASES, X-Y SYSTEM


The X-Y member stiffness can be derived in a similar manner to the X-Z
system. In general, only the sign of the moments will change. The resulting stiffness matrices for releasing Diz, qiy, Djz, and qjy are shown in Equations 5.7 through 5.10 with only the affected terms replaced with the new

Advanced Structural Stiffness 215

values. Equation 5.11 contains the stiffness matrix used when the member
has no stiffness contribution in the case of the cantilever beam, a pinnedpinned beam, or a pinned-slotted beam.

k1,1
0

0
0

k7 ,1
0

0
0

0
0
0
0
0

0
0
k3,3
0
k5,3

0
0
0
k 4 ,4
0

0
0
k3,5
0
k5,5

0
0
0
k10,4
0

0
0
k9,5
0
k11,5

0 0
0 0
0 k9,3
0 0
0 k11,3
0

0
0
0
0
0
EI z
L
0
0
0
0
0
EI z

k1,7
0
0
0
0

0
0
0
0
0

0
0
k3,9
0
k5,9

k4,10
0

k3,11
0
k5,11

0
0
0

0
0

k7 ,7
0
0
0
0
0

0 0
0 0
0 k9,9
0 0
0 k11,9
0

0
0
0

0
0

k10,10
0

k9,11
0
k11,11

0
0 ix Pix

0 iy Piy
P
0 iz iz
0 qix M ix

EI z qiy M iy

q
M
L iz iz
=

0 jx Pjx

0 jy Pjy

0 jz Pjz
q M
0 jx
jx
0 q jy M jy

EI z q jz M jz

(5.7)

0
k1,1

EI
3
z
0

L3
0
0

0
0

0
0

0
0

0
k7 ,1

3EI
0 3z
L

0
0

0
0
0
0

3EI z
0
L2

0 k1,7

0 0

k3,3 0 k3,5
0 k 4 ,4 0
k5,3 0 k5,5
0
0
0
0
0
0
0

0
0
0
0
0

0
0
0
0
k7 ,7

0 0

k9,3 0 k9,5 0 0
0 k10,4 0 0 0
k11,3 0 k11,5 0 0
0

0 0

0
3EI z
3
L
0
0
0
0
0
3EI z
L3
0
0
0
3EI
2z
L

k3,9 0 k3,11
0 k4,10 0
k5,9 0 k5,11
0
0
0
0
0
0
0

k9,9 0 k9,11
0 k10,10 0
k11,9 0 k11,11
0

0
3EI z
L2
0

0
0

0
3EI
2z
L
0

0
0
3EI z

ix Pix
P
iy iy
iz Piz


q ix M ix
q M
iy iy
q iz M iz
= P
jx jx
jy Pjy


jz Pjz
q M
jx jx
q jy M jy
q M
jz jz

(5.8)

216NUMERICAL STRUCTURAL ANALYSIS

k1,1
0

0
0

k7 ,1
0
0

0
0

0 0
0
0
0 0
0
0
0 k3,3 0 k3,5
0 0 k 4 ,4 0
0 k5,3 0 k5,5
0

0 0
0
0
0 0
0
0
0 k9,3 0 k9,5
0 0 k10,4 0
0 k11,3 0 k11,5
0

0 k1,7
0
0
0
0
0
0
0
0
EI z
0
L
0 k7 ,7
0
0
0
0
0
0
0
0
EI
z 0
L

0 0
0
0
0 0
0
0
0 k3,9 0 k3,11
0 0 k4,10 0
0 k5,9 0 k5,11
0

0
0 ix Pix

0 iy Piy
0 iz Piz

0 qix M ix
q

EI z iy M iy
q
M
L iz = iz

0 jx Pjx
0 jy Pjy
0 jz Pjz

0 q jx M jx
0 q jy M jy

EI z q jz M jz
L

0 0
0
0
0 0
0
0
0 k9,9 0 k9,11
0 0 k10,10 0
0 k11,9 0 k11,11
0

(5.9)
0
k1,1

3EI z
0
L3

0
0
0
0

0
0

3EI z
0
L2

k7 ,1
0

3
0 EI3 z

L
0
0

0
0
0
0
0
0

k3,3 0 k3,5
0 k 4 ,4 0
k5,3 0 k5,5
0

k9,3 0 k9,5
0 k10,4 0
k11,3 0 k11,5
0 0 0

0
k1,7
0
3EI z
3EI z
0 3
L2
L
0
0
0
0
0
0
0
0
0
3EI z
3EI
0 2z
L
L
0
k7 ,7
0
3EI
3EI z
2z 0
L
L3
0
0
0
0
0
0
0
0
0
0
0
0

k3,9 0 k3,11
0 k4,10 0
k5,9 0 k5,11
0

k9,9 0 k9,11
0 k10,10 0
k11,9 0 k11,11
0
0
0

0 ix Pix
iy Piy

0
iz Piz

0 q M
ix
ix
0 q M
iy
iy

0 q iz M iz
jx = Pjx
0 P
jy jy
0 jz Pjz
q M
0 jx jx
q M
0 jy jy
0 q jz M jz
0

(5.10)
k1,1
0

0
0

0
0
k
7 ,1
0
0

0
0

0 0
0 0
0 k3,3
0 0
0 k5,3
0 0
0 0
0 0
0 k9,3
0 0
0 k11,3
0 0

0
0
0
k 4 ,4
0
0
0
0
0
k10,4
0
0

0
0
k3,5
0
k5,5
0
0
0
k9,5
0
k11,5
0

0 k1,7
0 0
0 0
0 0
0 0
0 0
0 k7 ,7
0 0
0 0
0 0
0 0
0 0

0 0
0 0
0 k3,9
0 0
0 k5,9
0 0
0 0
0 0
0 k9,9
0 0
0 k11,9
0 0

0
0
0
k4,10
0
0
0
0
0
k10,10
0
0

0
0

k3,11
0
k5,11
0
0
0
k9,11
0
k11,11
0

0 ix Pix
0 iy Piy

0 iz Piz
0 qix M ix


0 qiy M iy
0 qiz M iz
=
0 jx Pjx

0 jy Pjy
0 jz Pjz


0 q jx M jx
0 q jy M jy


0 q jz M jz

(5.11)

Advanced Structural Stiffness 217

Table5.2. Release codesX-Y system


Djy

qiz

Djy

qjz

Equation

0
1
0
0
0
1
1
1
0
0
0
0
1
1
1
1

0
0
1
0
0
1
0
0
1
1
0
1
0
1
1
1

0
0
0
1
0
0
1
0
1
0
1
1
1
0
1
1

0
0
0
0
1
0
0
1
0
1
1
1
1
1
0
1

4.33
5.1
5.2
5.3
5.4
5.5
Unstable
5.5
5.5
5.5
5.5
Unstable
Unstable
Unstable
Unstable
Unstable

Table5.2 summarizes all the flexural stiffness conditions for the X-Y system. In this table, 1 indicates that the degree of freedom is released and 0
indicates that the degree of freedom is not released.

5.3 MEMBER END RELEASES, 3-D SYSTEM


When member releases occur in both coplanar systems, X-Z and X-Y, the
member stiffness matrix should be created by combining the appropriate
conditions from each system. The process for creating the global joint stiffness is the same as outlined in Section 4.13 with four modifications. First,
the released member stiffness matrix should be used for the appropriate
members in step one. Second, if a released member has a member load the
fixed-end forces and moments must be modified. These modified forces
and moments can be found in most structural analysis textbooks. Third, the
released member stiffness matrix should be used to solve for the local member forces in step six. Lastly, the modified fixed-end forces and moments
must be added to the local member end forces and moments in step six. The
following exampleuses a released member to model a structure.

218NUMERICAL STRUCTURAL ANALYSIS

Example5.3 Member stiffness


Determine the global joint deformations, support reactions, and local
member forces for the pin-connected bracing structure loaded as shown
in Figure5.5.
The area of each member, Ax, is 20 in2, the moment of inertia, Iz, is
1,000 in4, and the modulus of elasticity, E, is 10,000 ksi. Note the structure
is in the X-Y coordinated system and the numbering system is shown.
1.5 k/ft
4
8 ft

Hinge

3
X

2
3

12 ft

8 ft

12 ft

Figure5.5. Example5.3 Member stiffness.

The member stiffness for each member is first found from Equation 4.34
(step 1). Member 2 will have a released stiffness matrix and will be created last. Selecting member 2 instead of member 1 will result in not having
to use modified fixed-end forces and moments.
K g = [ R ] [ K m ][ R ]
T

K g = [a ] [ K m ][a ]
T

Using the local member stiffness for the X-Y system from Section 4.11,
which has translation in the x and y directions and rotation about the z direction, the rotation transformation of members will be about the z-axis or a.

[a ]T

cos -sin
sin cos

0
0
=
0
0
0
0

0
0

AE
L
0
0

0
0
0
0
0
0
1
0
0
0 cos -sin
0 sin cos
0
0
0

12 EI z
L3
6 EI z

6 EI z
L2
4 EI z

AE
L
0
0

0
0
0

0
0

1
0
12 EI z
L3
6 EI
2z

6 EI z
L2

2 EI z

[a ]T

cos -sin
sin cos

0
0
=
0
0
0
0

0
0

0
0
0
0
0
0
0
0
1
0
0
0

0 cos -sin 0
0 Advanced
sin cosStructural
0
Stiffness

0
0
0
1

AE
AE

0
0
0
L
L

12 EI z
6 EI z
12 EI
0
3 z
0

L2
L
L3

6
6
EI
4
EI
EI
z
z
0
2z
0

L
L2
L
[ K m ] = AE
AE

0
0
0
L
L

12 EI
6 EI
12 EI z
0
3 z 2z
0
L
L
L3

6 EI
6 EI z
2 EI z
2z
0
0
L

L2
L
0
0
0
cos sin 0
sin cos 0
0
0
0

0
0
1
0
0
0
[a ] = 0

0
0 cos sin 0

0
0
0 sin cos 0

0
0
0
0
1
0

219

6 EI z
L2

2 EI z
L

6 EI z
2
L
4 EI z

L
0

Members 1, 4, and 5 are already in the global system and do not need
rotation. Selecting the left end as the i-end, the rotation is 0. Tables 5.3
through 5.5 contain the local member stiffness matrices, the rotations
matrices, a and aT, and the global member stiffness matrices.
K 22

[K ] = K
21

12

K 44

[K ] = K
42

24

K 55

[K ] = K
53

35

K 21
K11
K 42
K 22
K 53
K 33

Member 3 is vertical and selecting the bottom end as the i-end, the rotation
is 90. Table5.6 contains the local member stiffness matrix, the rotations
matrices, a and aT, and the global member stiffness matrix.
K 33

[K ] = K
32

23

K 32
K 22

220NUMERICAL STRUCTURAL ANALYSIS

Table5.3. Example5.3 M
 ember stiffness, member 1
KL
1388.9

0.0

0.0

-1388.9

0.0

0.0

0.0

40.2

2893.5

0.0

-40.2

2893.5

0.0

2893.5

277777.8

0.0

-2893.5

138888.9

-1388.9

0.0

0.0

1388.9

0.0

0.0

0.0

-40.2

-2893.5

0.0

40.2

-2893.5

0.0

2893.5

138888.9

0.0

-2893.5

277777.8

a
1

aT
1

KG
1388.9

0.0

0.0

-1388.9

0.0

0.0

0.0

40.2

2893.5

0.0

-40.2

2893.5

0.0

2893.5

277777.8

0.0

-2893.5

138888.9

-1388.9

0.0

0.0

1388.9

0.0

0.0

40.2

-2893.5

-2893.5

277777.8

0.0

-40.2

-2893.5

0.0

0.0

2893.5

138888.9

0.0

For member 2, we will release the i-end at joint 1 for rotation. Selecting
the i-end at joint 1 the rotation is 213.7. Table5.7 contains the local member stiffness matrix, the rotation matrices, a and aT, and the global member stiffness matrix. The general released stiffness matrix is also shown.
K11

[K ] = K
13

31

K13
K 33

Advanced Structural Stiffness 221

Table5.4. Example5.3 Member stiffness, member 4


KL
833.3

0.0

0.0

-833.3

0.0

0.0

0.0

8.7

1041.7

0.0

0.0

1041.7

166666.7

0.0

-8.7
-1041.7

83333.3

-833.3

0.0

0.0

833.3

0.0

0.0

8.7

0.0

1041.7

-1041.7

0.0

-8.7
1041.7

0.0

83333.3

0.0

-1041.7

-1041.7
166666.7

1
0
0
0
0
0

0
1
0
0
0
0

0
0
1
0
0
0

0
0
0
1
0
0

0
0
0
0
1
0

0
0
0
0
0
1

1
0
0
0
0
0

0
1
0
0
0
0

0
0
1
0
0
0

0
0
0
1
0
0

0
0
0
0
1
0

0
0
0
0
0
1

833.3

0.0

0.0

-833.3

0.0

0.0

0.0

8.7

1041.7

0.0

-8.7

1041.7

0.0

1041.7

166666.7

0.0

-1041.7

83333.3

-833.3

0.0

0.0

833.3

0.0

0.0

0.0

-8.7
1041.7

-1041.7

0.0

8.7

83333.3

0.0

-1041.7

-1041.7
166666.7

aT

KG

0.0

AE
L

0
[ K m ] = AE

0
3EI z
L3
0
0
3EI z
L3
3EI z
L2

AE
L

AE
L

3EI z
L3
0
0

3EI z
L3
3EI
2z
L

3EI z
L2

3EI z
2
L
3EI z
L
0

222NUMERICAL STRUCTURAL ANALYSIS

Table5.5. Example5.3 Member stiffness, member 5


KL
1388.9

0.0

0.0

-1388.9

0.0

0.0

0.0

40.2

2893.5

0.0

-40.2

2893.5

0.0

2893.5

277778

0.0

-2893.5

138888.9

-1388.9

0.0

0.0

1388.9

0.0

0.0

0.0

-40.2

-2893.5

0.0

40.2

-2893.5

0.0

2893.5

138888.9

0.0

-2893.5

277778

aT
1

0.0

-1388.9

0.0

0.0

-40.2

2893.5

KG
1388.9

0.0

0.0

40.2

2893.5

0.0

0.0

2893.5

277778

0.0

-2893.5

138888.9

-1388.9

0.0

0.0

1388.9

0.0

0.0

40.2

-2893.5

-2893.5

277778

0.0

-40.2

-2893.5

0.0

0.0

2893.5

138888.9

0.0

The global joint stiffness matrix can be assembled using each of the members contributions (step 2). Table5.8 contains the global joint stiffness
matrix.
K11
K
21
K g = K 31

0
0

K12
K 22
K 32
K 42
0

K13
K 23
K 33
0
K 53

0
K 24
0
K 44
0

0
0
K 35

0
K 55

Advanced Structural Stiffness 223

Table5.6. Example5.3 Member stiffness, member 3


KL
2083.3

0.0

0.0

-2083.3

0.0

0.0

0.0

135.6

6510.4

0.0

-135.6

6510.4

0.0

6510.4

416666.7

0.0

-6510.4

208333.3

-2083.3

0.0

0.0

2083.3

0.0

0.0

0.0

-135.6

-6510.4

0.0

135.6

-6510.4

0.0

6510.4

208333.3

0.0

-6510.4

416666.7

a
6.12574E-17

-1

6.12574E-17

6.12574E-17

-1

6.12574E-17

aT
6.12574E-17

-1

6.12574E-17

6.12574E-17

-1

6.12574E-17

135.6

0.0

-6510.4

-135.6

0.0

-6510.4

0.0

2083.3

0.0

0.0

-2083.3

0.0

-6510.4

0.0

416666.7

6510.4

0.0

208333.3

-135.6

0.0

6510.4

135.6

0.0

6510.4

0.0

-2083.3

0.0

0.0

2083.3

0.0

-6510.4

0.0

208333.3

6510.4

0.0

416666.7

KG

The global joint loading is determined from Equation 4.35. In this case,
members 1 and 4 are loaded with a uniformly distributed load. The fixedend forces and moments due to the load must be calculated. Normally, the
fixed-end forces and moments are rotated into the global system before
they are placed in the global joint loading, but in this case the member is
already in the global system and no rotation is necessary (step 3). The load
matrix is in units of kips and inches (k-in).

224NUMERICAL STRUCTURAL ANALYSIS

Table5.7. Example5.3 Member stiffness, member 2


KL
1155.6

0.0

0.0

-1155.6

0.0

0.0

0.0

5.8

0.0

0.0

-5.8

1001.6

0.0

-1155.6

0.0
0.0

0.0
0.0

0.0
1155.6

0.0
0.0

0.0
0.0

0.0

-5.8

0.0

0.0

5.8

0.0

1001.6

0.0

0.0

-1001.6

-1001.6
173343.8

0
0

-0.832050294 -0.554700196
0.554700196 -0.832050294
0
0

0
0

1
0

-0.832050294 -0.554700196
0.554700196 -0.832050294

0
0.554700196

0
0

aT

-0.832050294 0.554700196
-0.554700196 -0.832050294
0
0

0
0

1
0

-0.832050294
-0.554700196 -0.832050294

KG
801.8
530.7

530.7
359.6

0.0
0.0

-801.8
-530.7

-530.7
-359.6

555.6
-833.4

0.0
-801.8
-530.7
555.6

0.0
-530.7
-359.6
-833.4

0.0
0.0
0.0
0.0

0.0
801.8
530.7
-555.6

0.0
530.7
359.6
833.4

0.0
-555.6
833.4
173343.8

FEP42 =

wL 0.125k/in ( 240in )
= 15k
=
2
2

wL2 0.125k/in ( 240in )


= 600k-in
=
12
12
wL 0.125k/in ( 240in )
= 15k
=
FEP24 =
2
2
2

FEM 42 =

0.125k/in ( 240in )
wL2
= 600k-in
=
FEM 24 =
12
12
wL 0.125k/in (144in )
FEP21 =
= 9k
=
2
2
2

wL2 0.125k/in (144in )


=
= 216k-in
12
12
wL 0.125k/in (144in )
FEP12 =
=
= 9k
2
2
2

FEM 21 =

wL2

0.125k/in (144in )

Advanced Structural Stiffness 225

Table5.8. Example5.3 Member stiffness


KG
2191

531

-1389

531

400

-2894

-40 -2894 -531 -360 -833

2894 138889

-1389
0
0

-802
-531
556

-2894 277778

-802 -531
0

556
0

0.125-k/in
) -833 0 0
wL
0
6510
0 in6510
136 ( 240
=
FEP42 =
= 15k
0
2132
0
2 1852 0 2 -2083 0
-40 2894
-9 -1042
2
2861111 -6510
0
208333
0
1042
83333
-2894 138889 6510 1852
0.125k/in ( 240in )
wL
= 600k-in
FEM 42 = 0
=
0
2326
531
0
-531
-136
-1389 0
12 -6510
12 -7066
0
0 -2083
0
531 2483 -2060
0
-360
-40 -2894
)
wL 0.125k/in ( 240in
=
=
=
FEP
15
k
0
6510
0
208333
867788
0
2894
138889
24
-833
-7066 -2060
2
2
0

2358

-833

833
2

2
0.125k/in ( 240in )0 9 1042
wL1042
9
= 600k-in
=

=
24
12
0 -1042 12
83333
0 1042 166667
144in )0
wL 0.125k/in
1389
-1389 ( 0
FEP21 =
= 9k
=
2
0 2 -40 2894
0

0
FEM

(
)
wL
=
= 216k-in
12
12
wL 0.125k/in (144in )
FEP12 =
=
= 9k
2
2
FEM 21 =

0 -2894
138889
0.125k/in
144in

0.125k/in (144in )
wL2
= 216k-in
=
12
12
2

FEM12 =

Pg = P & M g [ FEPM m ][ R ]

P1x 0
P 9
1y

M 1z 216

P2 x 0
P2 y 24

M 2 z 384
P 0
3x

= P3 y = 0

M 3z 0
P4 x 0

P4 y 15
M 600
4z

P5 x 0
P 0
5y

M 5 z 0

40

2894

2894 277778

226NUMERICAL STRUCTURAL ANALYSIS

The global deformations can be found from the global stiffness Equation 4.36 (step 4). The rows and columns corresponding to the support
constraint degrees of freedom must be deleted prior to the solution. This
would be all three motions at 4 and the X and Z motions at joint 5. The
resulting matrix is shown in Table5.9 along with the reduced load. The
solution for the deformations will be in inches and radians.
1x 0.07505
2.45212
1y

q1z 0.00004

2 x 0.06168
2.25618
1
g = K g Pg = 2 y =

q 2 z 0.00261
0.03701
3x

3 y 2.25254

q 3 z 0.00449
q 5 z 0.02122
The reactions at the supports can be found using the solution of the global
deformation with Equation 4.37 (step 5). Only the terms in the rows corresponding to the restrained degrees of freedom and in the columns of the
unrestrained degrees of freedom need to be included. Table5.10 shows the
appropriate stiffness terms and deformations needed to find the reactions.
Since there are fixed-end forces and moments at support joint 4, they must
Table5.9. Example5.3 Member stiffness
KG
2191

531

531

400

-1389
0
0

-FEPM

-1389

-531

556

-802

-2894

-40

-2894

-531

-360

-833

-9

-2894 277778

2894 138889

216

6510

-2083

-24

208333

384

2358

6510

-136

-40

2894

2132

1852

-2894 138889 6510 1852 861111 -6510

-802

-531

-531

-136

-360

-2083

556

-833

6510

-6510

2326

531

-7066

531

2483

-2060 -2894

208333 -7066 -2060 867788 138889

-2894 138889 277778

Advanced Structural Stiffness 227

Table 5.10 Example 5.3 Member stiffness


0 0 0 -833

0.0750

-51.4

-51.4

0 0 0

-9

1042

-2.4521

16.9

15

31.9

0 0 0

-1042 83333

0.0000

0 0 0

-1389

0.0617

51.4

51.4

0 0 0

-2.2562

16.1

16.1

-40 2894 2894

2132.9 600 2732.9

-0.0026
-0.0370
-2.2525
-0.0045

-0.0212

be added back to get the final reactions. The reaction forces are in kips and
inches (k-in).

[ P ] = K g g P & M g [ FEPM m ][ R ]T
[ P ] = K g g
The final step is finding the member forces for each of the members using
Equation 4.35 (step 6). The member force will be in kips and inches.
The local member stiffness matrix and the rotation matrix were shown
in step1 and are omitted here. The sign convention for the X-Y system
applies when interpreting the final end forces and moments.

[ P & M ] = [ K ][ R ]
[ P & M ] = [ K ][a ]
m

+ [ FEPM m ]

For member 1, the deformations at joints 2 and 1 are used. Since this
member had a load, the fixed-end forces and moments must be added to
the results. Table5.11 contains the final member end forces in the local
system along with global deformations used to find those end forces.
For member 2, the deformations at joints 1 and 3 are used. Table5.12
contains the final member end forces in the local system along with global
deformations used to find those end forces.
For member 3, the deformations at joints 3 and 2 are used. Table5.13
contains the final member end forces in the local system along with global
deformations used to find those end forces.

228NUMERICAL STRUCTURAL ANALYSIS

Table 5.11. Example 5.3 Member stiffness


DG

0.06
-2.26

0.00
0.08
-2.45

0.00

Dx2
Dy2

PL

Final P

-18.56

-18.56

0.45

9.45
64.27
18.56
8.55
0.00

-151.73

qz2

18.56

Dx1

-0.45

Dy1

216.00

qz1

kips
kips
kip-in
kips
kips
kip-in

Table 5.12. Example 5.3 Member stiffness


PL

DG

0.08
-2.45

0.00
-0.04
-2.25

0.00

20.19

Dx1

-3.18

Dy1

0.00

qz1

-20.19

Dx3

3.18

Dy3

-550.21

qz3

Final P
20.19

kips
kips
-3.18
0.00
kip-in
kips
-20.19
3.18
kips
-550.21 kip-in

Table 5.13. Example 5.3 Member stiffness


PL

DG

-0.04
-2.25

0.00
0.06
-2.26

0.00

Dx3
Dy3
qz3
Dx2
Dy2
qz2

7.58
-32.84
-1772.68
-7.58

32.84
-1379.89

Final P
7.58

kips
kips
-32.84
-1772.68 kip-in
kips
-7.58
32.84
kips
-1379.89 kip-in

For member 4, the deformations at joints 4 and 2 are used. Since this
member had a load, the fixed-end forces and moments must be added to
the results. Table5.14 contains the final member end forces in the local
system along with global deformations used to find those end forces.
For member 5, the deformations at joints 5 and 3 are used. Table5.15
contains the final member end forces in the local system along with global
deformations used to find those end forces.
Releases can also occur for the axial and torsional components of the
member stiffness. If either end of the member is released for Dx or qx, then
all the corresponding stiffness components are zero. If both ends of the

Advanced Structural Stiffness 229

Table 5.14. Example 5.3 Member stiffness


PL

DG

0.00
0.00
0.00
0.06

Dx4

-2.26

Dy2

0.00

Dy4
qz4
Dx2
qz2

-51.40

16.87
2132.90
51.40
-16.87

1915.62

Final P
kips
31.87
kips
2732.90 kip-in
51.40
kips
kips
-1.87
1315.62 kip-in
-51.40

Table 5.15. Example 5.3 Member stiffness


DG

0.00
0.00

Dx5

-0.02

qz5

-0.04
-2.25

0.00

Dy5
Dx3
Dy3
qz3

PL

Final P

51.40
16.13
0.00

51.40
16.13
0.00

-51.40
-16.13

2322.89

kips
kips
kip-in
-51.40 kips
-16.13 kips
2322.89 kip-in

member are released, then the member is unstable. The axial components
of the member stiffness matrix are k1,1, k1,7, k7,1, and k7,7. The torsional
components of the member stiffness matrix are k4,4, k4,10, k10,4, and k10,10.

5.4NON-PRISMATIC MEMBERS
Non-prismatic members have cross-sectional properties that vary along
the length of the member. The stiffness of these members can be handled
in two ways. First, the member could be divided into prismatic sections
and modeled with several different members of constant cross-section.
Second, the member can be modeled with stiffness derived from the mathematical model of the cross-sectional variation. The following derivation
is for the stiffness of a non-prismatic member in the X-Z system.
Example5.4 Non-prismatic member stiffness
Derive the local member stiffness in the X-Z system for a non-prismatic
cross-section using Castiglianos theorems.
Figure5.6 shows the beam with an applied deflection and an applied
rotation at the i-end. These can be used simultaneously to derive the stiffness of the member.

230NUMERICAL STRUCTURAL ANALYSIS


iz
Miy
Piz

Miy

iy

Piz

Figure5.6. Example5.4 Non-prismatic member stiffness.

The internal moment, Mx, at any point, x, can be found from statics
and the partial derivatives of that moment can be found with respect to the
applied force and moment at the i-end.
M x = Piz x M iy

dM x
= x
d Piz
dM x
= 1
d M iy
Castiglianos second theorem states that the first partial derivative of strain
energy with respect to a particular force is equal to the displacement of the
point of application of that force in the direction of its line of action. This
can be applied for both Diz and qiy.
iz = M x

d M x dx
dx
dx
= P x2
+ M x
d Piz EI y iz EI y iy EI y

qiy = M x

d M x dx
dx
dx
= Piz x
+ M iy
d M iy EI y
EI y
EI y

Since the cross-sectional properties vary, the moment of inertia, Iy, varies. Let the values S1, S2, and S3 be used and substituted into the previous
two equations. These values can be pre-derived for the cross-sectional
variation.
S1 =

dx
EI y

S2 = x

dx
EI y

S3 = x 2

dx
EI y

iz = Piz S3 + M iy S 2

qiy = Piz S2 + M iy S1

S1 =

dx
EI y

dx
S2 =Advanced
x EI y Structural Stiffness 231
S3 = x 2

dx
EI y

iz = Piz S3 + M iy S 2

qiy = Piz S 2 + M iy S1
Written in matrix form this is the flexibility matrix:

iz S3 S 2 Piz
q =

iy S 2 S1 M iy
[d i ] = [ f ii ][ Fi ]
We can solve this matrix equation for the forces Piz and Miy by the cofactor
method. Substituting the determinant of the flexibility matrix, D=S3S1
S22, to simplify the equation. This is the stiffness form of the equation.

Piz 1 S1
M =
iy D S 2
[ Fi ] = [ K ii ][d i ]

S 2 iz

S3 q iy

We can use the transmission matrix Equation 4.6 to find the forces at
the j-end of the member, where the values of x are cause minus effect or
x=xixj=0L=L.
1 0 1 0
=

x 1 L 1

[T ] =

K ji = [T ][ K ii ]
1 0 S1
K ji =

L 1 S 2

S 2 1 S1
S2 1
=

S3 D S 2 S1 L S 2 L S3 D
S 2 iz
Pjz 1 S1
M =

jy D S 2 S1 L S 2 L S3 iy
Fj = K ji [d i ]
The carry-over factor (COF) used in the moment distribution method can
be found by observing the ratio of the moment at the j-end to the moment
at the i-end.

232NUMERICAL STRUCTURAL ANALYSIS

COFi j

1
( S 2 S1 L ) iz + ( S 2 L S3 ) iy
D
=
=
1
M iy
S2 iz + S3iy
D
M jy

When only the rotational deformation is considered, the following is the


COF:
COFi j =

S 2 L S3
S3

The distribution factor (DF) used in the moment distribution method is the
ratio of the rotational stiffness of a member to the sum of the rotational
stiffness of all members at the joint. The rotational stiffness is the moment
at a joint due to the rotation at a joint. This is the term Kii for rotation and
moment only.
K Miy, q iy =

S3
D

The DF for a member at a joint can be written as follows:


S3
DF = D
S
D3

We could find the deflection and rotation at the j-end using the same
method. Alternatively, since the stiffness matrix is symmetric we can find
the forces at the i-end due to motions at the j-end directly.
T

S2
T
1 S1
1 S1
Kij = K ji =
= S

D S2 S1L S2 L S3
D 2
Piz 1 S1 S2 S1L jz
M =

iy D S2 S2 L S3 jy

S2 S1L
S2 L S3

Fi = Kij d i
We can use the transmission matrix equation again to find the forces at the
j-end of the member.

Advanced Structural Stiffness 233

K jj = [T ] K ij


1 0 S1
K jj =


L 1 S2
P jz 1 S1

=
M jy D S1L S2

Fj = K jj d j

S2 S1L 1 S1
=
S2 L S3 D S1L S2
S1L S2
jz

2
S1L 2 S2 L + S3 q jy

S1L S2

S1L 2 S2 L + S3 D
2

The COF from the j-end to the i-end is as follows considering the deformations at the j-end:

COF j i =

M iy
M jy

1
S2 jz + ( S2 L S3 ) q jy

( S L S ) + S L2 2 S L + S q 1
jz
3
jy
2
1
2
1
D

When only the rotational deformation is considered the following is the


COF:
COF j i =

S 2 L S3

S1L 2 S2 L + S3

The DF is a ratio of the rotational stiffness of a member to the sum of the


rotational stiffness of all members at the joint.
Rotational stiffness, the stiffness due to rotation at point A:
Kq B =

S1 L2 2 S 2 L + S3
D

The DF for a member at a joint can be written as follows:

DF =

S1L2 2 S2 L + S33
D
S1L2 2S2 L + S3
D

234NUMERICAL STRUCTURAL ANALYSIS

The axial stiffness terms can be derived directly from strength of materials. The axial stiffness is the inverse of the flexibility, which can be written
as follows. The torsional stiffness would look the same as the axial stiffness with GIx substituted for EAx:

f =

dx

EA

The coplanar X-Z nonprismatic stiffness matrix is shown in Equation 5.12.


The X-Y system will have the same values with the sign of the moments
due to deflection reversed.
1
f

0
[ Km ] = 1

S1
D
S
2
D

S2
D
S3
D

S1
D
S 2 S1L
D

S2
D
S 2 L S3
D

1
f

0
0
1
f
0
0

0
S1
D
S2
D

0
S1
D
S1L S2
D

S 2 S1L

S 2 L S3

S1L S 2

2
S1L 2 S2 L + S3

D
0

(5.12)
FIXED-END MOMENTS
The fixed-end forces and moments must be derived for a non-prismatic
member. The changes in stiffness along the length of the member will
change how the forces and moments are distributed by the member.
The following examples derive two of the most common member
loads.
Example5.5 Non-prismatic member stiffness
Derive the fixed-end forces and moments due to a uniformly distributed load in the X-Z system for a non-prismatic cross-section using
Castiglianos theorems.

Advanced Structural Stiffness 235


W

FEMiy

FEPiz

FEMjy

FEPjz
L

Figure5.7. Example5.5 Non-prismatic member stiffness.

The free-body diagram of the beam is shown in Figure5.7. The procedure for Castiglianos second theorem used in Example5.4 will be
repeated here.
The internal moment, Mx, at any point, x, can be found from statics
and the partial derivatives of that moment can be found with respect to the
applied force and moment at the i-end.
Mx =

wx 2
FEPiz x FEM iy
2

dM x
= x
d FEPiz
dM x
= 1
d FEM iy
iz = 0 = M x

d M x dx
dx
dx
wx3 dx
= FEPiz x 2
+ FEM iy x

d FEPiz EI y
EI y
EI y
2 EI y

qiy = 0 = M x

d M x dx
dx
dx
wx 2 dx
= FEPiz x
+ FEM iy

EI y
EI y
d FEM iy EI y
2 EI y

Observing that there is a new term that varies with x, we will substitute S4
as follows:
S 4 = x 3

dx
EI y

w
S4
2
w
0 = FEPiz S 2 + FEM iy S1 S3
2
S3 S 2 FEPiz S 4 w
S S FEM = S 2
iy
2
1
3
0 = FEPiz S3 + FEM iy S 2

FEPiz 1 S1
FEM =
D S2
iy

S2 S4 w
w S1S 4 S 2 S3
=

S3 S3 2 2 D S32 S 2 S 4

236NUMERICAL STRUCTURAL ANALYSIS

Apply equilibrium on Figure5.7 to find the j-end forces and moments.


FEPjz = wL FEPiz
FEM jy =

wL2
FEPiz L FEM iy
2

Example5.6 Non-prismatic member stiffness


Derive the fixed-end forces and moments due to a concentrated load in
the X-Z system for a non-prismatic cross-section using Castiglianos
theorems.
The free-body diagram of the beam is shown in Figure5.8. The procedure used in Example5.5 will be repeated here.
FEMiy

FEPiz

FEMjy

FEPjz

Figure5.8. Example5.6 Non-prismatic member stiffness.

The internal moment, Mx, at any point, x, can be found from statics
and the partial derivatives of that moment can be found with respect to the
applied force and moment at the i-end. In this case, two moment equations
must be written. The first is Mx1, with x from the i-end to the point load
(0 x a) and the second is Mx2, from the point load to the j-end (a x L).
M x1 = FEPiz x FEM iy
M x 2 = P ( x a ) FEPiz x FEM iy
The partial derivatives are the same for either of the two moment equations.

dM x
= x
d FEPiz
dM x
= 1
d FEM iy
iz = 0 = M x

qiy = 0 = M x

L
d M x dx
dx
dx
dx
= FEPiz x 2
+ FEM iy x
Px ( x a )
d FEPiz EI y
EI y
EI y a
EI y
L
d M x dx
dx
dx
dx
= FEPiz x
+ FEM iy
P ( x a )
d FEM iy EI y
EI y
EI y a
EI y

Advanced Structural Stiffness 237

Observing that there are two new terms that vary with x, we will substitute
S5 and S6 as follows:
S5 = ( x a )

dx
EI y

S6 = ( x 2 ax )

dx
EI y

0 = FEPiz S3 + FEM iy S 2 PS6


0 = FEPiz S 2 + FEM iy S1 PS5
S3 S 2 FEPiz S6
S S FEM = S P
iy
2
1
5
EPiz 1 S1 S 2 S6
FE
P S S S 2 S5
P= 1 6
FEM =

D S 2 S3 S5
D S3 S5 S 2 S6
iy

Apply equilibrium on Figure5.8 to find the j-end forces and moments.


FEPjz = P FEPiz
FEM jy = Pb FEPiz L FEM iy
The process for solving stiffness problems involving non-prismatic members is the same as with prismatic members with the same four modifications for members with end releases. The following exampleshows this
process.
Example5.7 Non-prismatic member stiffness
Determine the deformations at the free end of the non-prismatic beam
using only three degrees of freedom in the stiffness solution. Also determine the final end forces and the support reactions. The beam is shown in
Figure5.9.
Z

A=10 in2 A=20 in2


I=100 in4 I=200 in4
W=600 lb/ft

X
100in

100in

E=10,000 ksi

Figure5.9. Example5.7 Non-prismatic member stiffness.

238NUMERICAL STRUCTURAL ANALYSIS

S1 =

dx
=
EI y

S2 = x

100

dx
=
EI y

200

dx
dx
1 100 0 200 100
+
=
+

= 0.00015
100 E 100 200 E 10000 100
200

100

200
1 1002 02 2002 1002
xdx
xdx
+
=
+

2 ( 200 )
100 E 100 200 E 10000 2 (100 )

= 0.0125
S3 = x 2

dx
+
EI y

100

x 2 dx
1 1003 03 2003 1003
=
+

= 1.5
200 E 10000 3 (100 )
3 ( 200 )
100
200

x 2 dx

100 E +
0

D = S1 S3 S 2 = 0.00015 (1.5 ) ( 0.0125 ) = 0.00006875


2

100

200

dx 100 dx
dx 200 dx
dx
10000 200
200100
100
11 100
dx
f ==
.0015
==
++
==
10 ++
==00.0015
10EE 100
20
E
10000
20
20
10000
10
20
EEAx x 00 10
E

100
K ==

11
11
666.67
.67
==
==666
f 00.0015
.0015

Since the j-end of the member is fixed, there is no need to build the entire
member stiffness matrix. The j-end motions will be eliminated and only
the i-end of the member stiffness needs to be developed.

1
f

[ Kii ] = 0

0
S1
D
S
2
D

0
666.67
0
0
S2
= 0
2.1818 181.81
D
0
181.81 21818
S3
D

The fixed-end forces and moments must be derived and applied to the
system.
S4 = x 3

dx
EI y

100

200
x 3 dx
x 3 dx
1 1004 04 2004 1004
+
+
=

= 212.5
4 ( 200 )
100 E 100 200 E 10000 4 (100 )

0
0

w
Pg = [ P ] [ FEPM ] = FEPiz =
( S1S 4 S 2 S3 )
2D

2
FE

EM iy
( S3 S 2 S 4 )

0
0.05

Pg =
0.00015 ( 212.5 ) 0.0125 (1.5 ) = 4.7727

2 ( 0.00006875 )
2

147.73

(1.5 ) 0.0125 ( 212.5 )

Advanced Structural Stiffness 239

The general stiffness equation can be set-up and solved as follows. The
units are in inches and radians:

[ Pi ] = [ Kii ][d i ]
1
[d i ] = [ Kii ] [ Pi ]
0
0
666.67

2.1818 181.81
[d i ] = 0
0
181.81 21818

0 0
4.7727 = 5.3123

147.73 0.0375

5.5 SHEAR STIFFNESS, X-Z SYSTEM


The shear stiffness of a member should be included when it is significant.
This effect was developed by Timoshenko (1921) in 1921. For normal
frame structures, the stiffness contributions due to shear are minor and
are sometimes ignored. For frames and structures with larger or deeper
members, the shear stiffness contribution is appreciable and should be
included. The following two sections and corresponding examples derive
the combined flexural and shear stiffness of members.
Example5.8 Shear stiffness
Derive the local member shear stiffness for qiy using Castiglianos
theorems.
The free-body diagram of the beam with an imposed rotation of qiy is
shown in Figure5.10. Also shown is a left-hand free-body of the beam cut
at any distance x from the i-end.
Z
Miy

Mjy
X
iy
Pjz

Piz
L

Miy

M
Piz

Figure5.10. Example5.8 Shear stiffness.

240NUMERICAL STRUCTURAL ANALYSIS

The internal shear, Vx, and moment, Mx, at any point, x, can be found
from statics and the partial derivatives of that shear and moment can be
found with respect to the applied force and moment at the i-end.
Vx = Piz

dVx
= 1
d Piz
dVx
=0
d M iy
M x = Piz x M iy

dM x
= x
d Piz
dM x
= 1
d M iy
Castiglianos second theorem can be applied noting that at the i-end the
deflection is zero and the rotation is qiy.
iz = 0 = M x
0=

d M x dx
d Vx dx
dx
dx
+ Piz
+ Vx
= ( Piz x 2 + M iy x )
d Piz EI y
d Piz GAz
EI y
GAz

2
Piz L3 M iy L Piz L
+
+
3EI y 2 EI y GAz

M iy
EI y

2 Piz L 2 Piz

3EI y LGAz

q iy = M x
q iy =

d M x dx
d Vx dx
dx
dx
+ 0
+ Vx
= ( Piz x + M iy )
d M iy EI y
d M iy GAz
GAz
EI y

Piz L2 M iy L
+
EI y
2 EI y

Substituting the first equation into the second equation results in the following:

q iy =

Piz L2 2 Piz L2 2 Piz


P L2 2 P

= iz iz
2 EI y 3EI y GAz
6 EI y GAz

L2
2

q iy =
Piz
6 EI y GAz
6 EI y Az G
Piz = 2
q iy
L Az G + 12 EI y

(5.13)

Advanced Structural Stiffness 241

Substituting Equation 5.13 into the following equation repeated from earlier, results in Equation 5.14 for the second stiffness value.
M iy
EI y

2 Piz L
P
iz
3EI y LGAz

4 EI y L2 Az G + 3EI y
M iy =
L3 Az G + 12 LEI y

) q

iy

(5.14)

Example5.9 Shear stiffness


Derive the local member shear stiffness for Diz using Castiglianos
theorems.
The free-body diagram of the beam with an imposed deflection of Diz
is shown in Figure5.11. Also shown is a left-hand free-body of the beam
cut at any distance x from the i-end.
Z
Miy
Mjy
X

iz

Piz

Pjz

Miy

M
Piz

Figure5.11. Example5.9 Shear stiffness.

The internal shear, Vx, and moment, Mx, are exactly the same as in
Example5.8. The partial derivatives are also the same.
Vx = Piz

dVx
= 1
d Piz
dVx
=0
d M iy
M x = Piz x M iy

dM x
= x
d Piz
dM x
= 1
d M iy

Vx = Piz

dVx
= 1
d Piz
dVx
=0
d M iy

242NUMERICAL STRUCTURAL ANALYSIS

M x = Piz x M iy

dM x
= x
d Piz
dM x
= 1
d M iy
Castiglianos second theorem can be applied noting that at the i-end the
rotation is zero and the deflection is Diz.

qiy = 0 = M x
0=

dVx dx
d M x dx
dx
dx
+ Vx
= Piz x + M iy
+ 0
d M iy EI y
d M iy GAz
EI y
GAz

Piz L2 M iy L
+
EI y
2 EI y

Piz L
2
d M x dx
dVx dx
dx
dx
iz = M x
= Piz x 2 + M iy x
+ Piz
+ V
EI y GAz
d Piz EI y x d Piz GAz
M iy =

iz =

2
Piz L3 M iy L Piz L
+
+
3EI y 2 EI y GAz

Substituting the first equation into the second equation results in the
following:

iz =

Piz L3 Piz L3 Piz L


P L3
PL

+
= iz
+ iz
3EI y 4 EI y GAz 12 EI y GAz

L3
L
+
iz =
Piz
12E
EI y GAz
12 EI y Az G

Piz = 3
L A G + 12 LEI iz
z
y

(5.15)

Substituting Equation 5.15 into the following equation repeated from earlier, results in Equation 5.16 for the second stiffness value.
Piz L
2
6 EI y Az G
M iy = 2
iz
L Az G + 12 EI y
M iy =

(5.16)

Advanced Structural Stiffness 243

The forces at the j-end of the member due to the motions at the i-end can
be found using the transmission matrix. Then, the forces at the i-end due
to the motions at the j-end can be found by symmetry of the stiffness
matrix. Finally, the forces at the j-end due to motions at the j-end can be
found using the transmission matrix. This process was illustrated previously in Chapter 4. The resulting terms are shown in the matrices given as
Equations 5.21 through 5.24.

5.6 SHEAR STIFFNESS, X-Y SYSTEM


The following contains the combined flexural and shear stiffness of members in the X-Y system.
Example5.10 Shear stiffness
Derive the local member shear stiffness for qiz using Castiglianos
theorems.
The free-body diagram of the beam with an imposed rotation of qiz is
shown in Figure5.12. Also shown is a left-hand free-body of the beam cut
at any distance x from the i-end.
Y
Miz

iz

Piy

Mjz
X

Pjy

Miz

M
Piy

Figure5.12. Example5.10 Shear stiffness.

The internal shear, Vx, and moment, Mx, at any point, x, can be found
from statics and the partial derivatives of that shear and moment can be
found with respect to the applied force and moment at the i-end.

244NUMERICAL STRUCTURAL ANALYSIS

Vx = Piy

dVx
= 1
d Piy
dVx
=0
d M iz
M x = Piy x M iz
dM x
=x
d Piy
dM x
= 1
d M iz
Castiglianos second theorem can be applied noting that at the i-end the
deflection is zero and the rotation is qiz.

iy = 0 = M x
0=

Piy L3
3EI z

d M x dx
dV dx
dx
dx
+ Vx x
= Piy x 2 M iz x
+ Piy
d Piy EI z
d Piy GAy
EI z
GAy

M iz L2 Piy L
+
2 EI z GAy

M iz 2 Piy L 2 Piy
=

EI z 3EI z LGAy
iz = M x
iz =

d M x dx
dVx dx
dx
dx
+ Vx
= Piy x + M iz
+ 0
d M iz EI z
d M iz GAy
EI z
GAy

Piy L2
2 EI z

M iz L
EI z

Substituting the first equation into the second equation results in the
following:

q iz =

Piy L2
2 EI z

2 Piy L2
3EI z

2 Piy
GAy

Piy L2
6 EI z

2 Piy
GAy

L2
2
Piy
+
q iz =
EI z GAy
6E
6 EI z Ay G
Piy = 2
q iz
L Ay G + 12 EI z

(5.17)

Advanced Structural Stiffness 245

Substituting Equation 5.17 into the following equation repeated from earlier, results in Equation 5.18 for the second stiffness value.
M iz 2 Piy L 2 Piy
=

EI z 3EI z LGAy

4 EI z L2 Ay G + 3EI z
M iz =
L3 Ay G + 12 LEI z

) q

(5.18)

iz

Example5.11 Shear stiffness


Derive the local member shear stiffness for Diy using the Castiglianos
theorems.
The free-body diagram of the beam with an imposed deflection of Diy
is shown in Figure5.13. Also shown is a left-hand free-body of the beam
cut at any distance x from the i-end.
Y
Mjz

Miz

iy

Piy

Pjy

Miz

M
Piy

Figure5.13. Example5.11 Shear stiffness.

The internal shear, Vx, and moment, Mx, are exactly the same as in
Example5.10. The partial derivatives are also the same.
Vx = Piy

dVx
= 1
d Piy
dVx
=0
d M iz
M x = Piy x M iz
dM x
=x
d Piy
dM x
= 1
d M iz

Vx = Piy

dVx
= 1
d Piy
dVx
=0
dM
246NUMERICAL STRUCTURAL izANALYSIS
M x = Piy x M iz
dM x
=x
d Piy
dM x
= 1
d M iz
Castiglianos second theorem can be applied noting that at the i-end the
rotation is zero and the deflection is Diy.

qiz = 0 = M x
0=
M iz =

Piy L2
2 EI z

M iz L
EI z

Piy L
2

iy = M x
iy =

d M x dx
dVx dx
dx
dx
+ Vx
= Piy x + M iz
+ 0
d M iz EI z
d M iz GAy
EI z
GAy

Piy L3
3EI z

d M x dx
dV dx
dx
dx
= Piy x 2 M iz x
+ Piy
+ Vx x
d Piy EI z
d Piy GAy
EI z
GAy

M iz L2 Piy L
+
2 EI z GAy

Substituting the first equation into the second equation results in the
following:
iy =

Piy L3
3EI z

Piy L3
4 EI z

Piy L
GAy

Piy L3
12 EI z

Piy L
GAy

L3
L
Piy
+
iy =
EI z GAy
12E
12 EI z Ay G
Piy = 3
iy
L Ay G + 12 LEI z

(5.19)

Substituting Equation 5.19 into the following equation repeated from earlier, results in Equation 5.20 for the second stiffness value.
M iz =

Piy L
2

6 EI z Ay G
M iz = 2
iy
L Ay G + 12 EI z

(5.20)

Advanced Structural Stiffness 247

The resulting terms are shown in the matrices given as Equations 5.21
through 5.24.

5.7SHEAR STIFFNESS, 3-D SYSTEM


The total three-dimensional (3-D) shear stiffness matrices are shown in
four parts in the following text as Equations 5.21 through 5.24. These
parts represent the force at the i-end or j-end due to the motions at the
i-end or j-end. This is shown in the total stiffness matrix as follows:
K

K ij
K jj

[ K ] = K ii

Ax E
0
0
L

12 EI z Ay G

0
0 L3 A G + 12 LEI
y
z

12 EI y Az G
0
0
3
L
A

z G + 12 LEI y
[ Kii ] =
0
0
0

6 EI y Az G
0
0
2

L Az G + 12 EI y

6 EI z Ay G
0
0

L2 Ay G + 12 EI z

ji

0
I xG
L
0
0

6 EI y Az G
L2 Az G + 12 EI y
0

4 EI y L2 Az G + 3EI y
L3 Az G + 12 LEI y
0

6 EI z Ay G

L2 Ay G + 12 EI z

2
4 EI z L Ay G + 3EI z

L3 Ay G + 12 LEI z
0

(5.21)

Ax E
0
0
0
0
0
L

12 EI z Ay G
6 EI z Ay G

0
0
0
2
0 L3 A G + 12 LEI

2
EI
+
1
L
A
G
z
y
z
y

12 EI y Az G
6 EI y Az G
0

3
2
0
0
0
L Az G + 12 LEI y
L Az G + 12 EI y

K ji =
I
G
0

x
0
0
0
0

2
2 EI y ( L Az G + 6 EI y )
6 EI y Az G

0
0
0
0

L2 Az G + 12 EI y
L3 Az G + 12 LEI y

2
+
EI
L
A
G
EI
2
6
EI
6
A
G
z (
y
z )
z y

0
0
0
0

L2 Ay G + 12 EI z
L3 Ay G + 12 LEI z

(5.22)

248NUMERICAL STRUCTURAL ANALYSIS

K ij =

Ax E
L
0

12 EI z Ay G
L3 Ay G + 12 LEI z

0
0

12 EI y Az G

I xG
L

6 EI y Az G

L2 Az G + 12 EI y

6 EI z Ay G

L2 Ay G + 12 EI z

L3 Az G + 12 LEI y

6 EI y Az G
L2 Az G + 12 EI y
0

2 EI y L Az G 6 EI y
L3 Az G + 12 LEI y

6 EI z Ay G

2
L Ay G + 12 EI z

2
2 EI z L Ay G 6 EI z
L3 Ay G + 12 LEI z
0

(5.23)

Ax E
0
0
L

12 EI z Ay G

0
0 L3 A G + 12 LEI
y
z

12 EI y Az G
0
0
L3 Az G + 12 LEI y

K jj =
0
0
0

6 EI y Az G
0
0

L2 Az G + 12 EI y

6 EI z Ay G
0
0

L2 Ay G + 12 EI z

0
I xG
L
0

6 EI y Az G
L2 Az G + 12 EI y
0

4 EI y L2 Az G + 3EI y
L3 Az G + 12 LEI y

6 EI z Ay G

2
L Ay G + 12 EI z

4 EI z L2 Ay G + 3EI z

L3 Ay G + 12 LEI z
0

(5.24)
5.7.1 SHEAR AREA
The shear area is the cross-sectional property that is used for shear energy
resistance. It can be found for a cross-section using the shear stress equation for a beam derived in most strength of materials textbooks. The basic
equations for shear in the Y and Z direction are given as follows:

t xy =

Vy Qy
I z tz

Vy Ay '

I z tz
V Q V Az '
t xz = z z = z
I yt y
I yt y

Advanced Structural Stiffness 249

If these equations are written using a single term to represent all the
cross-sectional properties, it results in the following:

t xy =
t xz =

Vy Qy
I z tz

Vy Ay '
I z tz

Vy
Ay

Vz Qz Vz Az ' Vz
=
=
I yt y
I yt y
Az

The equations for the shear areas Ay and Az can then be found.

Ay =

I z tz
A y '

Az =

I yt y
Az '

There are three terms in these equations. The first is the centroidal moment
of inertia. The second is the moment of the area, Ay or Az, between the
centroid and the extreme fiber taken about the centroid. The third is the
value t, which is the width at the centroid.
Example5.12 Shear area
Determine the shear areas Ay and Az for the a rectangular section.
Figure5.14 shows the rectangle in the orientation to calculate Az that
corresponds to bending about the Y axis. To find Ay the area to the left of
the Z axis will be used.

y h

Figure5.14. Example5.12 Shear area.

250NUMERICAL STRUCTURAL ANALYSIS

hb3
h
It
2bh
Ay = z z = 12 =
bh b
A y '
3
2 4
bh3
b
I yt y
2bh
Az =
= 12 =
Az ' bh h
3
2 4
For a rectangle, the shear areas are both two-thirds the cross-sectional
area.
Example5.13 Shear area
Determine the shear area Az for the T-shaped section shown in Figure5.15.
The centroid and the centroidal moment of inertia are found using the
moment of area principles.
z=

Az
A

2 (12 ) 6 + 12 ( 2 )13
2 (12 ) 2

= 9.50 in

2 (12 )
12 ( 2 )
2
2
+ 2 (12 ) ( 9.5 6 ) +
+ 12 ( 2 ) ( 9.5 13)
12
12
= 884.0 in 4
3

I y = I y + Ad z2 =

12"
z
2"
y
12"

2"

Figure5.15. Example5.13 Shear area.

Advanced Structural Stiffness 251

The area of the web below the centroid will be used to find Az.
Az =

I yty
Az '

884.0 ( 2 )
= 19.59 in 2
9.5
2 ( 9.5 )
2

Table5.16 shows the calculations in tabular form.


Table 5.16. Example 5.13 Shear area
Element
Chord
Web

b
12
2

h
z
2
13.0
12
6
z = 9.5
Iy = 884.00

A
Az
24
312
24
144
48
456
A'z' = 90.25

I
8
288
296

d
3.5
3.5

Ad2
294
294
588
Az = 19.590

5.8GEOMETRIC STIFFNESS, X-Y SYSTEM


An ordinary stiffness analysis, whether it includes shear deformations or
not, makes no adjustments for the changing geometry of a loaded structure. Forces and moments are calculated from the original positions of
the joints, not from their deformed positions. Elastic buckling, which is
a function of joint deformations, is therefore impossible to predict using
ordinary stiffness analysis. A procedure to include member and joint deformations in force and moment calculations can be developed by assuming
a deformed shape and calculating the additional moment such as deformation would cause. Figure5.16 shows a member subjected to bending
and axial force in an un-deformed and deformed shape. An alternate, yet
similar, derivation is published by Ketter, Lee, and Prawel (1979).
The first is a beam in Figure5.16 that was used to derive the ordinary
elastic stiffness matrix in Section 4.11 for the X-Y system. In that case, the

Miz
Pix

Pix
iy
Miz
Piy

Figure5.16. Geometric stiffness.

Piy
L

252NUMERICAL STRUCTURAL ANALYSIS

moment in the beam is a function of only the end shears and moments, as
given by the following equation:
M x = Piy x M iz
The second beam in Figure5.16 shows the bending deformations. In this
case, the internal moment is a function of not only the end shears and
moments, but also a function of the axial force multiplied by the beams
lateral deflection, y.
M x = Piy x M iz + Pix y
This additional moment, the product of axial force, Pix, and lateral deflection, y, is usually called the P-delta effect. To derive a stiffness matrix
that includes the P-delta effect, equilibrium of the deformed beam must
be considered.
Example5.14 Geometric stiffness
Derive the Diy stiffness using Castiglianos theorems for a linear member
including the geometric effects.
Using the principle of superposition, consider a beam with an applied
deflection while the rotation is held to zero. Figure 5.17 shows the
deformed beam with applied end forces. Also shown is a left-hand freebody of the beam cut at any distance x from the i-end.
The internal bending moment in the beam is found from equilibrium.

M x = Piy x M iz + Pix iy y
Miz
Pix
iy

Piy

L
Miz
Pix
iy-y

M
P
Piy

V
x

Figure5.17. Example5.14 Geometric stiffness.

Advanced Structural Stiffness 253

If the lateral deformation, y, is assumed to be a general cubic function,


the four known boundary conditions can be used to find the particular
solution.
y = ax 3 + bx 2 + cx + d
y ( 0) = iy
y ( L) = 0

y ' ( 0) = 0

y ' ( L) = 0
2 x3 3x 2
y = iy 3 2 + 1
L
L

Substituting this equation into the internal moment equation yields the
following:
2 x3 3x 2
M x = Piy x M iz + Pix iy 3 + 2
L
L
It is assumed that axial shortening is caused only by the axial force, Pix.
This is the same assumption used for the ordinary elastic stiffness derivation. The geometric stiffness derivation considers the lateral and rotational
deformations, Diy and qiz. From Castiglianos theorem, the general deflection and rotation of the free end are as follows:

iy = M x

d M x dx
d Piy EI z

qiz = M x

d M x dx
d M iz EI z

The partial derivatives of the internal moment equation with respect to


applied shear force and moment are the following:

dM x
=x
d Piy
dM x
= 1
d M iz

254NUMERICAL STRUCTURAL ANALYSIS

It should be noted that rotation and deflection are functions of moment


only, since shear deformations are ignored in this derivation. Setting qiz to
zero and solving for Piy and Miz in terms of the deflection will result in two
terms of the stiffness matrix.

qiz = 0 =

M iz = Piy

L
1
+ Pix iy
2
2

2 x 3 3 x 2 dx
Piy x + M iz + Pix iy 3 2
L EI z
L

L
L
L2
+ M iz
Pix iy
0 = Piy
2 EI z
2 EI z
EI z

2 x 4 3 x 3 dx
2
Piy x M iz x + Pix iy 3 + 2
L EI z
L

7 L2
L3
L2
+ Pix iy
iy = Piy
M iz
10 EI z
3EI z
2 EI z
iy =

Substituting the first equation into the second equations yields the
following:

iy = Piy

L3
L2
+ Pix iy
12 EI z
10 EI z

Piy = iy

12 EI z
6
+ Pix iy
3
5L
L

(5.25)

Substituting Equation 5.25 into the following equation repeated from earlier results in Equation 5.25 for the second stiffness value.
1
L
+ Pix iy
2
2
6 EI
1
(5.26)
M iz = iy 2 z + Pix iy
10
L
M iz = Piy

Take note that the first terms in each of these stiffness equations are the
same as the elastic stiffness values derived in Equations 4.26 and 4.27.
The second term is the geometric component due to the deflected shape
and the axial thrust.

Advanced Structural Stiffness 255

Example5.15 Geometric stiffness


Derive the qiz stiffness using Castiglianos theorems for a linear member
including the geometric effects.
In this case, consider a beam with a known rotation while the deflection is held to zero. Figure5.18 shows the deformed beam with applied
end forces. Also shown is a left-hand free-body of the beam cut at any
distance x from the i-end.
Miz
Pix
Piy

iz
L

Miz
Pix

M
Piy

iz

Figure5.18. Example 5.15 Geometric stiffness.


The internal bending moment in the beam is found from equilibrium.
M x = Piy x M iz Pix y

If the lateral deformation, y, is assumed to be a general cubic function,


the four known boundary conditions can be used to find the particular
solution.

y = ax 3 + bx 2 + cx + d
y ( 0) = 0

y ( L) = 0

y ' ( 0) = qiz
y ' ( L) = 0

x3 2 x 2

y = qiz 2
+ x
L
L

x3 2 x 2

x
M x = Piy x M iz + Pix qiz 2 +
L
L

256NUMERICAL STRUCTURAL ANALYSIS

The general deflection and rotation of the free end are the same as
Example5.14.

iy = M x

d M x dx
d Piy EI z

qiz = M x

d M x dx
d M iz EI z

The partial derivatives of the internal moment equation with respect to


applied shear force and moment are the same as in Example5.14.
M x
=x
d Piy

d Mx
= 1
d M iz
Setting Diy to zero and solving for Piy and Miz in terms of the deflection will
result in two terms of the stiffness matrix.

x3 2 x 2
dx
+ x
Piy x + M iz + Pix qiz 2
L
L

EI z

L
L2
L2
+ M iz
+ Pix qiz
qiz = Piy
2 EI z
EI z
12 EI z

qiz =

EI
L
L
Pix qiz qiz z
2
12
L

x 4 2 x3
dx
iy = 0 = Piy x 2 M iz x + Pix qiz 2 +
x2
L
L

EI z

M iz = Piy

0 = Piy

L2
L3
L3
M iz
Pix qiz
3EI z
2 EI z
30 EI z

Substituting the first equation into the second equation yields the following:
Piy = q iz

6 EI z
1
+ Pixq iz
2
10
L

(5.27)

Advanced Structural Stiffness 257

Substituting Equation 5.27 into the following equation repeated from earlier results in Equation 5.28 for the second stiffness value.

EI
L
L
Pix qiz qiz z
2
12
L
4 EI z
2 L

M iz = qiz
+ Pix qiz
L
15

M iz = Piy

(5.28)

Take note that the first terms in each of these stiffness equations are the
same as the elastic stiffness values derived in Equations 4.24 and 4.25.
The second term is the geometric component due to the deflected shape
and the axial thrust. All four terms can be written in matrix form.
12 EI z
L3

6 EI z
2
L

6 EI z
6
2
5L
L
+ Pix
4 EI z
1
10
L

1
10 iy Piy
=

2 L qiz M iz

15

The first matrix on the left side of the equation in the basic elastic stiffness
matrix will be called [K]. The second matrix on the left side of the equation in the geometric stiffness matrix will be called [G]. In general terms,
the equation may be written as follows:

([ K ] + P [G ]) [ ] = [ F ]
ix

The same transformation used for previous stiffness matrix derivations can
be applied to find the rest of the geometric stiffness matrix. The geometric
stiffness matrix for the coplanar X-Y system is given as Equation 5.29.
The sign convention on Pix is positive for tension.

0
0

6
0
5L

1
0
10
[ K m ] = Pix 0 0

0 6

5L

1
0

10

0
1
10
2L
15
0
1

10
L

30

0
6
5L
1

10
0
6
5L
1

10

0
0
0
0
0

0
1
10
L

30 
0

1

10

2L
15

(5.29)

258NUMERICAL STRUCTURAL ANALYSIS

5.9GEOMETRIC STIFFNESS, X-Z SYSTEM


The geometric stiffness matrix for the coplanar X-Z system can be
derived in a similar manner to the X-Y system performed in Section 5.8.
The primary difference is that the signs of the moments due to translation and the forces due to rotation will be the opposite of Equation 5.29.
The geometric stiffness matrix for the coplanar X-Z system is given as
Equation 5.30.
0
0

6
0
5L

1
0
10
[ K m ] = Pix 0 0

0 6

L
5L

1
0

10

1
10
2L
15
0

1
10
L

30

0
6
5L
1
10
0
6
5L
1
10

0
0
0
0
0

0
1

10
L

30
0

1
10

2L
15

(5.30)

Example5.16 Geometric stiffness


Determine the deformations at the free end of the beam by including both
the elastic and geometric stiffness contributions to the stiffness solution.
The beam is shown in Figure5.19.
Since the j-end of the member is fixed, there is no need to build the
entire member stiffness matrix. The j-end motions will be eliminated and

Z
50kips
1kip

A=10 in2
I=100 in4
200in
E=10,000 ksi

Figure5.19. Example5.16 Geometric stiffness.

Advanced Structural Stiffness 259

only the i-end of the member stiffness needs to be developed. The value of
Pix in this case is known to be 50 kips.
EAx
L

0
12 EI z
L3
6 EI
2z
L

0 0

6 EI
6
2 z + Pix 0

5L
L

4 EI z
0 1
10
L

0
ix Pix
1

iz = Piz
10
q M
2 L iy iy
15

500
0
0 0 0
0 ix 50

0
1.5
150 + 0 .3
5 iz = 1

0 150 20, 000 0 5 1333.3 q 0


iy

0
0
ix 50
500

0
1.2
145 iz = 1

0 145 18, 666.7 qiy 0


Solving the general solution as shown in the following equations, the
resulting deformations can be found. The solution is in inches and
radians.

([ K ] + P [G ]) [ ] = [ F ]
[ ] = ([ K ] + P [G ]) [ F ]
ix

ix

ix 0.1000

iz = 13.5757
qiy 0.10545

If the solution was performed with the geometric stiffness omitted, the
result would be as follows:
ix 0.1000

iz = 2.6667
qiy 0.02000

260NUMERICAL STRUCTURAL ANALYSIS

Example5.17 Geometric stiffness


Determine the deformations at the free end of the beam by including both
the elastic and geometric stiffness contributions to the stiffness solution.
The beam is shown in Figure5.20.
Z
100kips
1kip

A=10 in2
I=100 in4

200in
E=10,000 ksi

Figure5.20. Example5.17 Geometric stiffness.

This is the same as Example5.16, except the axial force has been
increased to 100 kips.
500
0
0 0 0
0 ix 100

0
1.5
150 + 0 .6
10 iz = 1

0 150 20, 000 0 10 2666.7 q 0


iy

0
0
ix 100
500

0
0.9
140 iz = 1

0 140 17, 333.3 qiy 0


ix 0.2000

iz = 4.33333
qiy 0.00350

Observe that the i-end of the beam moved in the negative Z direction. This
does not make logical sense. The reason of the backward motion is that the
member has buckled elastically. The actual elastic buckling load of this
column is 61.685 kips.

5.10GEOMETRIC STIFFNESS, 3-D SYSTEM


By combining Equations 5.29 and 5.30 and adding the torsional stiffness
terms, the geometric stiffness matrix in the 3-D Cartesian coordinate system can be found.

Advanced Structural Stiffness 261

0
Pix
0

0
6
5L
0
0
0
1
10
0
6

5L
0
0
0
1
10

6
1
0
5L
10
0 0 0
1
2L

0
10
15
0

6
1
0
5L
10
0 0 0
L
1

0
10
30

0 0 0
1
6
0
10
5L

1
2L
0
15
10
0 0 0
1
6

0
10
5L

0
0

L
1
0
30
10

6
1

0
5L
10
0 0 0
L
1
0
10
30
0

6
1
0
5L
10
0 0 0
1
2L
0
10
15
0

0
1
10

0
0

L

30
0
1

10
0

2L

15

ix Pix
iy Piy


iz Piz
q ix M ix
q M
iy iy
q iz M iz
jx = Pjx
P
jy jy
jz Pjz
q M
jx jx
q jy M jy
q M
jz jz

(5.31)

The geometric stiffness of a member can also be derived based on a general transcendental equation. The full derivation is published by Blette
(1985).
np
np
y = asin + bcos + cx + d
2L
2L

The particular solution for Diy is as follows:

px
px px p
2
y = iy
+

sin
+ cos
4 p
2L
2L 2L 2
The particular solution for qiz is as follows:

px
px px
4 L p
1 sin
y = qiz
+ cos
+
1
2
4p p 2
2L
2 L 2 L
These two relationships can be used to develop the geometric stiffness
matrix.

262NUMERICAL STRUCTURAL ANALYSIS


0

0
Pix
0

0
0

0
1.2036
L
0
0
0
0.1018
0
1.2036

0.1018

1.2036
L
0
0.1018
0
0
0

1.2036
0

L
0
0
0
0.1018
0.1018
0

0
1.2036
0
L

0 0.1018

0
0
0 0.1379 L
0
0
0
0

0
0
0.1379 L
0

0
0
0
0

0.1018 0

0.1018

0
0
0
0
0 0.0361L
0
0
0
0
0.0361L 0

1.2036
0

L
0
0
0
0.1018
0
0.1018
0
0
1.2036
0
L
1.2036
0
L
0
0
0
0.1018
0
0.1018

0
0
0
0
0
0
0
0
0
0

0
0.1018 ix Pix

iy Piy


0.1018
0
iz Piz


0
0
M
ix ix

0.0361L
0
M
iy
iy

0
0.0361L iz M iz
jx = Pjx
0
0


P

0
0.1018 jy jy

jz Pjz

jx
M jx
0.1018
0

M
jy
jy
0
0
M

jz
jz

0.1379 L
0

0
0.1379 L

(5.32)

If more digits are desired for accuracy, the following substitutions can be made:
1.2036=1.20362445
0.1018=0.1018122226
0.1379=0.1378809597
0.0361=0.03606873710

5.11GEOMETRIC AND SHEAR STIFFNESS


The effect of both the geometric and shear stiffness could be included in
the flexural stiffness derivations. This matrix was derived by Karl J. Blette
(Blette 1985). The procedure to derive the stiffness matrix would be to
include the shear stiffness contribution used in Sections 5.5 and 5.6 in
the geometric stiffness of Sections 5.8 and 5.9. Equation 5.33 shows the
elastic and shear member stiffness in the 3-D Cartesian coordinate system.
Equation 5.34 shows the elastic geometric and shear member stiffness in
the 3-D Cartesian coordinate system.
The terms a and b are defined as follows:

ay =

L3
12 EI y

az =

L3
12 EI z

by =

L
GAz

bz =

L
GAy

Ax E
L

Ax E
L

L
2 (a z + b z )

L
2 (a y + b y )

1
a y + by

1
a z + bz

IxG
L

L
2 (a y + b y )

I xG
L

L
2 (a z + b z )

1
a y + by

1
a z + bz

L
2 (a y + b y )

L (a y + b y )

EI y ( 2 a y b y )

L
2 (a y + b y )

L (a y + b y )

EI y ( 4 a y + b y )

L
2 (a z + b z )

L (a z + b z )

EI z ( 2 az b z )

L (a z + b z )

EI z ( 4 az + b z )

L
2 (a z + b z )

Ax E
L

Ax E
L

1
a z + bz

L
2 (a z + b z )

1
a z + bz

2 (a z + b z )

1
a y + by

L
2 (a y + b y )

I xG
L

IxG
L

1
a y + by

L
2 (a y + b y )

L (a y + b y )

EI y ( 4 a y + b y )

L
2 (a y + b y )

L (a y + b y )

L
2 (a y + b y )

EI y ( 2 a y b y )

2 (a z + b z )

P
0
ix
Piy
Piz
0
M ix

EI z ( 2 a z b z ) M iy
M
L (a z + b z ) = iz
Pjx
P
0
jy
Pjz
L
M

2 (a z + b z ) jx
M
jy
0
M jz

EI z ( 4 a z + b z )
L (a z + b z )
(5.33)
0

Advanced Structural Stiffness 263

0
0
0
( 6a z + 5b z )

5L ( a z + b z )

0
0

0
0

0
0

L
0

2 (a z + b z )

0
0

0
( 6a z + 5b z )

5L ( a z + b z )

0
0

0
0

0
0

az
0

a
10
( z + bz )

0
0

( 6a y + 5b y )
5L (a y + b y )

0
ay

10 a y + b y

0
0

0
ay

( 6a y + 5b y )
5L ( a y + b y )

10 a y + b y

0
0

0
0

)
60 (a y + b y )

)
60 (a y + b y )

L 2a y + 5b y

10 a y + b y

ay

L 8a y + 5b y

10 a y + b y

ay

0
0

60 (a z + b z )

0
0

0
az
10 (a z + b z )

60 (a z + b z )

L ( 8a z + 5b z )

L ( 2a z + 5b z )

0
0

0
az

10 (a z + b z )

0
0

0
0

0
0

0
0

10 a y + b y

0
ay

( 6a y + 5b y )
5L ( a y + b y )

( 6a z + 5b z )
5L (a z + b z )

0
0

10 a y + b y

az
10 (a z + b z )

0
ay

( 6a y + 5b y )
5L (a y + b y )

0
0

az
10 (a z + b z )

0
0

5L (a z + b z )

( 6a z + 5b z )

0
0

0
0

0
0

)
60 (a y + b y )

10 a y + b y

ay

L 8a y + 5b y

)
60 (a y + b y )

L 2a y + 5b y

10 a y + b y

ay

0
0

(5.34)

L ( 2a z + 5b z )
60 (a z + b z )

az

10 (a z + b z )

L ( 8a z + 5b z )

60 (a z + b z )

0
az

10 (a z + b z )

264NUMERICAL STRUCTURAL ANALYSIS

Advanced Structural Stiffness 265

5.12TORSION
The torsional stiffness of slender linear members is composed of two parts.
The first is known as St. Venants torsion, which is uniform on a member
at any distance, r, from the longitudinal axis. This is the torsional stiffness
that is included in the elastic member stiffness Equation 4.33. This is the
primary resistance to torsion for circular crosses-sections that have area
distributed uniformly about the longitudinal axis. The second type of torsional stiffness is known as warping torsion. Warping torsion is the primary
stiffness in thin-walled open cross-sections such as angles shapes and wide
flange shapes. The warping torsion will cause longitudinal deformations in
the cross-section that will cause certain portions to elongate and other portions to shorten. This warping effect can be included in the derivation of the
stiffness. Structural Analysis and Design, by Ketter, Lee, and Prawel, Jr.,
covers this derivation (Ketter, Lee, and Prawel 1979). The torsional stiffness
at each degree of freedom is represented as two components instead of the
single values used in the normal elastic stiffness. Equation 5.35 shows the
elastic torsional member stiffness in the 3-D Cartesian coordinate system.
Ax E
0
0
L

12 EI z
0
0

L3

12
EI

y
0
0
L3

0
0
0
0
0
0

6 EI y

2
0
0
L

6 EI z
0
0

L2

A
E
x
0
0
L

12 EI z

0
0 L3

12 EI y

3
0
0
L

0
0
0
0
0
0

6 EI y

2
0
0
L

6 EI z
0
0
L2

6 EI z
L2

T1 T2
T2 T3

6 EI y
L2
0
0

4 EI y

0
0

0
0

4 EI z
L

T1 T2
T2 T4
0
0

0
0

L
0

12 EI z
L3

0
0

L2
0
0
2 EI y

6 EI y

Ax E
L

12 EI y
L3
0
0
6 EI y

0
0

T1 T2
T2 T4

6 EI y
L2
0
0

2 EI y

12 EI z
L3

0
0

0
0

0
0

6 EI
2z
L

6 EI z
L2

2 EI y
L

Ax E
L

6 EI z
L2

L2

12 EI z
L3
0
0
6 EI y
L2
0

T1 T2
T2 T3
0

6 EI y
L2
0
0
4 EI y
L
0

6 EI z
2
L

0
0

2 EI z

6 EI
2z
L

0
0

4EI z

L
0

(5.35)

266NUMERICAL STRUCTURAL ANALYSIS

The resulting member stiffness matrix is 1414 in size with the terms T1,
T2, T3, and T4 defined as follows:
T1 = GT
T2 = GT

l sinh ( l L )

2 cosh ( l L ) 1 l Lsinh ( l L )
cosh ( l L ) 1

2 cosh ( l L ) 1 l Lsinh ( l L )

sinh ( l L ) l Lcosh ( l L )
GT
l 2 cosh ( l L ) 1 l Lsinh ( l L )
l L l sinh ( l L )
GT
T4 =
l 2 cosh ( l L ) 1 l Lsinh ( l L )
T3 =

l=

GT
EIw

In the equation for l, kT is the St. Venant torsion constant, which is typically the polar moment of inertia, Ix. The value of Iw is known as the warping constant. Both of these values are normally tabulated in handbooks or
specifications.

5.13SUB-STRUCTURING
When a structure is of large enough size that the contents for the global
joint stiffness matrix cannot be contained in the RAM of a computer, the
matrix can be transformed into segments by reduction or decomposition.
The resulting transformed matrix can take many forms depending on the
process used. One of the common transformations is the N-matrix, due to
the configuration of resulting values. The following is a general description of the operation used to solve large systems using the N-matrix. This
method can be used by operating on the individual degrees of freedom or
on the entire joint as matrix operations.
The original equation set is normally a sparse matrix with most of
the values near the main diagonal. Equation 4.34 represents the original
stiffness solution set. For clarity, the zero values are left out of the matrices
and X indicates where values exist.
K g g = Pg

Advanced Structural Stiffness 267

X
X

X
X
X
X

X
X
X

X
X
X
X

=
FEPM

X
X
X


(5.36)

The matrix is reduced by the appropriate reduction method. The degrees of


freedom or joints used to represent the stiffness should be placed at the outer
ends of the matrix. These will be the degrees of freedom in the outer columns of Equation 5.37. All of the values above and below the main diagonal
of the matrix will be reduced to zero except those retained in the outer columns. The resulting matrix is shown in Equation 5.37 and is the N-matrix.

Kg
X

X
X

K g

X
X
X
X

K g g Pg
X d S
X d A
= 
X d V
X d E

K g g Pg

(5.37)

The information contained in the N-matrix is used in two ways. The values
in the extreme boundaries of the matrices represent the stiffness, deformation, or force (Kg, Dg, and Pg) of the structure related to the retained degrees
of freedom or joints. These values can be combined with other members
in a stiffness analysis. The interior portions of the matrices represent the
values of the stiffness, deformation, or force of the degrees of freedom or
joints eliminated in the reduction. These values are used to find the deformations and forces of these degrees of freedom or joints once the main analysis
results are known. The N-matrix can be compressed as shown in Equation
5.38. The interior stiffness values can be inverted for later processing.

Kg
X

X
X

K g

X
X 1
X 1
X 1

K g g Pg
X S
X A
=
X V
X E

K g g Pg

(5.38)

268NUMERICAL STRUCTURAL ANALYSIS

The main analysis is performed and the global deformation of the retained
degrees of freedom of joint is found.
g = K g

Pg

The interior loads to be used to find the remainder of the deformations


are shown in Equation 5.39. This represents the loads due to the interior
loading that was saved and the load due to the motion of the main analysis
motions on the interior joints.

L S X

O A X
= +
A V X
D E X

g

X
X

X
X

g

(5.39)

The deformations of the interior degrees of freedom or joints can be found


from Equation 5.40.


X 1 L


X 1 O
= 1
X A
X 1 D



(5.40)

The local member forces are found from the deformation in Equation
5.40, the same as in step 6 of the general stiffness procedure given in
Section 4.13. The values of D are used for Dg in Equation 4.38 and
repeated here.
P & M = K [R ] + FEPM
m m
m

REFERENCES
Blette, K.J. 1985. Numerical Methods for Architectural Engineers. Stillwater,
Oklahoma: Oklahoma State University.

Advanced Structural Stiffness 269

Ketter, R.L., G.C. Lee, and S.P. Prawel. 1979. Structural Analysis and Design.
New York, New York: McGraw-Hill Book Company.
Timoshenko, S.P. 1921. On the Correction Factor for Shear of the Differential
Equation for Transverse Vibrations of Bars of Uniform Cross-section. Philosophical Magazine 41, pp. 7446.

About the Authors


Steven OHara is a Professor of Architectural Engineering and Licensed
Engineer who has taught architectural engineering at the Oklahoma State
University School of Architecture since 1988. His primary areas of interest include the design and analysis of masonry, steel and timber structures, with special interest in classical numerical structural analysis and
the design of concrete structures. Professor OHara is one of four faculty
members in the Architectural Engineering program at OSU, and as teaches
courses in the AE program at all levels; he enjoys his close mentoring relationship with the students in the AE program at OSU, as he also performs
the role of their academic advisor. Outside the OSU classroom, Professor
OHara is an Affiliate Professor of Civil Engineering and Architecture in
Project Lead the Way, a nationwide program for high-school students.
He trains the high school teachers responsible for introducing engineering
principles into the secondary curriculum, through project-based learning
and has coauthored the workbook for the curriculum Civil Engineering &
Architecture Workbook. He is the coauthor of ARE Review Manual used
by thousands of architects to prepare for their licensing exams.
As a faculty member, Professor OHara has received numerous
awards: In 2013, he was elected to the first class of Faculty Teaching Fellows at Oklahoma State University and was elevated to Chapter Honor
Member by Chi Epsilon. In 2001, he was the honored recipient of the first
Melvin R. Lohmann Professorship. In 1999, he received the Chi Epsilon
Excellence in Teaching Award. In 1995, he was recognized as the Halliburton Excellent Young Teacher for the College of Engineering, Architecture, and Technology at OSU.
Carisa H. Ramming joined the faculty at Oklahoma State University
in January 2009. Professor Ramming is a graduate of Oklahoma State
University with a bachelor of architectural engineering degree and dual
masters degrees: master of science in civil engineering with an emphasis
in Construction Engineering and master of architectural engineering. Professor Ramming teaches engineering mechanics: statics, steel design, and
building foundations.

272 About the Authors

Before returning to academics, Professor Ramming worked as a


consulting engineer primarily designing large structural steel retail facilities, and obtained her license as a professional engineer in the state of
Oklahoma.
Professor Ramming is a member of the American Society of Engineering Educators and the Oklahoma Structural Engineers Association.
She has served the engineering community as Director, Vice-President,
President, and Delegate of Oklahomas Chapter of the National Council
of Structural Engineering. Professor Ramming is the advisor for OSUs
student chapter of the Architectural Engineering Institute; the College of
Engineering, Architecture and Technologys Student Council; and Tau
Beta Pi.
Professor Ramming recently coauthored Civil Engineering & Architecture Workbook for Project Lead the Way, which provides curriculum
for science, technology, engineering, and mathematics education used in
middle and high schools. She was also named the Outstanding Faculty
Member from the College of Engineering, Architecture, and Technology
by The Panhellenic Council and Interfraternity Council in 2010 and 2012.
In 2013, she was awarded the Halliburton Young Teaching Award and in
2014, she received the Excellent Teacher Award for the College of Engineering, Architecture and Technology.

Index
A
Adjoint matrix, 60, 6263
Algebraic equation, 1
Alpha rotation matrix, 148149
Area moment method, 155158
B
Bairstows method, 3846
Basket weave method, 5556
Beta rotation matrix, 149150
Bisection method, 1315
C
Carry-over factor, 171
Castiglianos second theorem,
165168
Cholesky decomposition method,
7378
Cofactor matrix, 57, 59, 61
Column buckling, difference
operators, 134136
Column matrix, 49
Conjugate beam method, 158161
Coordinate systems, 147148
Cramers rule
definition, 56
eigenvalues by, 8790
example, 57
Crouts method, 7378
D
Descartes rule, 36
Diagonal matrix, 4950

Difference operators
column buckling with,
134136
fixed beam with, 128133
partial differential equations,
136140
simple beam with, 123128
Taylor series expansion,
118123
Distribution factor, 171
Double integration
Gaussian quadrature, 114115
Simpsons one-third rule,
112114
E
Eigenvalues, Cramers rule, 8790
Elastic member stiffness
three-dimensional (3-D)
system, 186187
X-Y system, 181186
X-Z system, 174181
Equations
algebraic, 1
homogeneous, 47
linear algebraic, 47
non-homogeneous, 47
simultaneous, 4748
transcendental, 12
Error equations, 7880
F
FaddeevLeverrier method, 9091

274Index

False position method, 1517


Fixed beam with difference
operators, 128133
Fixed-end moments, 234239
Flexibility, 174
Flexural stiffness, 171
G
Gamma rotation matrix, 150152
Gaussian elimination method,
6367
Gaussian quadrature method
description, 109112
double integration, 114115
GaussJordan elimination method
description, 6871
improved, 7273
GaussSeidel iteration method,
8687
Geometric stiffness
flexural stiffness derivations,
262264
three dimensional (3-D)
system, 260262
X-Y system, 251257
X-Z system, 258260
Global joint stiffness, 187203
Graeffes root squaring method
description, 2427
real and complex roots, 3538
real and distinct roots, 2831
real and equal roots, 3135
H
Homogeneous equation, 47
I
Identity matrix, 50
Improved GaussJordan
elimination method,
7273
Incremental search method, 1012
Inverted matrix, 51
Iterative method, 9194

J
Joint stiffness factor, 171
L
Linear algebraic equations, 47
Linear interpolation method,
1517
Lower triangular matrix, 50
M
Matrix
adjoint, 60, 6263
cofactor, 57, 59, 61
column, 49
definition, 48
diagonal, 4950
identity, 50
inverted, 51
minor, 51
row, 49
square, 49
transposed, 51
triangular, 50
Matrix factorization, 7378
Matrix inversion method, 8086
Matrix operations
matrix addition and
subtraction, 5253
matrix determinants, 5456
matrix multiplication, 5354
scalar multiplication, 53
Matrix structural analysis
area moment method, 155158
Castiglianos second theorem,
165168
conjugate beam method,
158161
global joint stiffness, 187203
moment-distribution method,
170174
slope-deflection method,
168170
transformation matrix
rotation matrix, 148152

Index 275

transmission matrix,
153155
virtual work method, 161163
visual integration method,
163165
Member end releases
three dimensional (3-D)
system, 217229
X-Y system, 214217
X-Z system, 205214
Member relative stiffness factor,
171
Member stiffness factor, 171
Method of false position, 1517
Moment-distribution method,
170174
N
NewtonRaphson method, 1821
Newtons second order method,
2124
Newtons tangent method, 1821
Non-homogeneous equation, 47
Non-prismatic member stiffness,
229234
Numerical integration and
differentiation
difference operators
column buckling with,
134136
fixed beam with, 128133
partial differential
equations, 136
140
simple beam with, 123128
Taylor series expansion,
118123
double integration
Gaussian quadrature,
114115
Simpsons one-third rule,
112114
Gaussian quadrature method,
109112

partial differential equations,


numeric modeling,
140145
Rombergs integration method,
99104
Simpsons rule, 104109
Taylor series polynomial
expansion, 116118
trapezoidal rule, 9799
O
Orthogonal forces, 153
P
Partial differential equations
difference operators, 136140
numeric modeling, 140145
Partial pivoting, 63
Pivot coefficient/element, 63
Pivot equation, 63
Polynomials, 23
Power method, 9194
R
Real and complex roots, 3538
Real and distinct roots, 2831
Real and equal roots, 3135
Refined incremental search
method, 1213
Rombergs integration method,
99104
Root methods
Bairstows method, 3846
bisection method, 1315
Descartes rule, 36
false position method, 1517
Graeffes root squaring method
description, 2427
real and complex roots,
3538
real and distinct roots,
2831
real and equal roots,
3135

276Index

incremental search method,


1012
linear interpolation method,
1517
NewtonRaphson method,
1821
Newtons second order method,
2124
Newtons tangent method,
1821
refined incremental search
method, 1213
secant method, 1718
synthetic division, 710
Rotation matrix
alpha, 148149
beta, 149150
gamma, 150152
Row matrix, 49
S
Scalar multiplication, 53
Secant method, 1718
Shear area, 248251
Shear stiffness
flexural stiffness derivations,
262264
shear area, 248251
three dimensional (3-D)
system, 247248
X-Y system, 243247
X-Z system, 239243
Simple beam with difference
operators, 123128
Simpsons one-third rule, 112114
Simpsons rule, 104109
Simultaneous equations, 4748
Simultaneous linear algebraic
equation methods
Cholesky decomposition
method, 7378
eigenvalues, Cramers rule,
8790
error equations, 7880

FaddeevLeverrier method,
9091
GaussJordan elimination
method, 6871
GaussSeidel iteration method,
8687
improved GaussJordan
elimination method,
7273
iterative method, 9194
matrix inversion method,
8086
matrix operations, 5256
power method, 9194
Slope-deflection method, 168170
Square matrix, 49
Stiffness
definition, 174
elastic member
three-dimensional (3-D)
system, 186187
X-Y system, 181186
X-Z system, 174181
flexural, 171
geometric
flexural stiffness
derivations,
262264
three dimensional (3-D)
system, 260262
X-Y system, 251257
X-Z system, 258260
global joint, 187203
shear
flexural stiffness
derivations,
262264
shear area, 248251
three dimensional (3-D)
system, 247248
X-Y system, 243247
X-Z system, 239243
torsion, 265266
Stiffness factor, 171

Index 277

Sub-structuring, 266268
Synthetic division, 710
T
Taylor series polynomial
expansion, 116118
Three-dimensional (3-D) system
elastic member stiffness,
186187
geometric stiffness, 260262
member end releases, 217229
shear stiffness, 247248
Torsional stiffness, 265266
Transformation matrix
rotation matrix
alpha, 148149
beta, 149150
gamma, 150152
transmission matrix
cause and effect, 153155
orthogonal forces, 153
Transcendental equation, 12
Transmission matrix
cause and effect, 153155
orthogonal forces, 153

Transposed matrix, 51
Trapezoidal rule, 9799
Triangular matrix, 50
U
Upper triangular matrix, 50
V
Virtual work method, 161163
Visual integration method,
163165
X
X-Y system
elastic member stiffness,
181186
geometric stiffness, 251257
member end releases, 214217
shear stiffness, 243247
X-Z system
elastic member stiffness,
174181
geometric stiffness, 258260
member end releases, 205214
shear stiffness, 239243

FORTHCOMING TITLES IN OUR SUSTAINABLE


STRUCTURAL SYSTEMS COLLECTION
Mohammad Noori, Editor
System Dynamics Modeling of Community Development Projects
By Bernard Amadei
Momentum Press is one of the leading book publishers in the field of engineering,
mathematics, health, and applied sciences. Momentum Press offers over 30 collections,
including Aerospace, Biomedical, Civil, Environmental, Nanomaterials, Geotechnical,
and many others.
Momentum Press is actively seeking collection editors as well as authors. For more
information about becoming an MP author or collection editor, please visit
http://www.momentumpress.net/contact

Announcing Digital Content Crafted by Librarians


Momentum Press offers digital content as authoritative treatments of advanced engineering topics by leaders in their field. Hosted on ebrary, MP provides practitioners, researchers, faculty,
and students in engineering, science, and industry with innovative electronic content in sensors
and controls engineering, advanced energy engineering, manufacturing, and materials science.
Momentum Press offers library-friendly terms:





perpetual access for a one-time fee


no subscriptions or access fees required
unlimited concurrent usage permitted
downloadable PDFs provided
free MARC records included
free trials

The Momentum Press digital library is very affordable, with no obligation to buy in future years.
For more information, please visit www.momentumpress.net/library or to set up a trial in the US,
please contact mpsales@globalepress.com.

Create your own


Customized Content
Bundlethe more
books you buy,
the greater your
discount!

Manufacturing
Engineering
Mechanical
& Chemical
Engineering
Materials Science
& Engineering
Civil &
Environmental
Engineering
Electrical
Engineering

THE TERMS
Perpetual access for
a one time fee
No subscriptions or
access fees
Unlimited
concurrent usage
Downloadable PDFs
Free MARC records
For further information,
a free trial, or to order,
contact:
sales@momentumpress.net

Steven OHara and Carisa H. Ramming


As structural engineers move further into the age of digital computation and rely more heavily on computers to solve problems,
it remains paramount that they understand the basic mathematics and engineering principles used to design and analyze building structures. The link between the basic concepts and application to real world problems is one of the most challenging
learning endeavors that structural engineers face.
The primary purpose of Numerical Structural Analysis is to
assist structural engineering students with developing the ability to solve complex structural analysis problems. This book will
cover numerical techniques to solve mathematical formulations,
which are necessary in developing the analysis procedures for
structural engineering. Once the numerical formulations are understood, engineers can then develop structural analysis methods that use these techniques. This will be done primarily with
matrix structural stiffness procedures. Finally, advanced stiffness
topics will be developed and presented to solve unique structural problems, including member end releases, non-prismatic,
shear, geometric, and torsional stiffness.
Steven OHara is a professor of architectural engineering and licensed engineer at the Oklahoma State University School of Architecture since 1988. Professor OHara is an affiliate professor
of civil engineering and architecture in Project Lead the Way,
a nationwide program for high-school students. He trains the
high school teachers responsible for introducing engineering
principles into the secondary curriculum, through project-based
learning and coauthored the workbook for the curriculum, Civil
Engineering & Architecture Workbook. He is also coauthor of
the ARE Review Manual.
Carisa H. Ramming is a graduate of Oklahoma State University
with a bachelor of architectural engineering degree and dual
masters degree: master of science in civil engineering with an
emphasis in construction engineering and master of architectural engineering. Professor Ramming teaches engineering mechanics and has been faculty at OSU since 2009. She recently
coauthored Civil Engineering & Architecture Workbook for
Project Lead the Way, which provides curriculum for science,
technology, engineering, and mathematics education used in
middle and high schools.
ISBN: 978-1-60650-488-8

Numerical Structural Analysis

THE CONTENT

Numerical Structural Analysis

OHARA RAMMING

EBOOKS
FOR THE
ENGINEERING
LIBRARY

SUSTAINABLE STRUCTURAL
SYSTEMS COLLECTION
Mohammad Noori, Editor

Numerical
Structural
Analysis

Steven OHara
Carisa H. Ramming